25 MB lightweight cross-platform database client for 100+ databases, including MySQL, PostgreSQL, SQLite, Redis, MongoDB, DuckDB, SQL Server, and Dameng. Built-in AI, MCP Server, CLI, desktop and Docker. | 轻量级跨平台数据库管理工具,支持 MySQL、PostgreSQL、SQLite、Redis、MongoDB、达梦等 100+ 数据库,提供桌面端、Docker、CLI、内置 AI 助手和 MCP。
TypeScript aicliclickhousedatabasedatabase-clientdatabase-managementdockerguimcpmongodbmysqlpostgresqlredisrustsql-serversqlitetaurivue
use serde::{Deserialize, Serialize};
use sqlparser::dialect::OracleDialect;
use sqlparser::tokenizer::{Token, Tokenizer};
use crate::models::connection::DatabaseType;
pub const MIN_FUZZY_FILTER_CHARS: usize = 2;
pub fn fuzzy_filter_enabled(filter: &str) -> bool {
filter.trim().chars().count() >= MIN_FUZZY_FILTER_CHARS
}
pub fn fuzzy_subsequence_match(text: &str, filter: &str) -> bool {
let filter = filter.trim().to_lowercase();
if filter.is_empty() {
return true;
}
let text = text.to_lowercase();
let mut chars = text.chars();
for needle in filter.chars() {
if !chars.any(|candidate| candidate == needle) {
return false;
}
}
true
}
pub fn contains_or_fuzzy_match(text: &str, filter: &str) -> bool {
let filter = filter.trim().to_lowercase();
if filter.is_empty() {
return true;
}
let text = text.to_lowercase();
text.contains(&filter) || (fuzzy_filter_enabled(&filter) && fuzzy_subsequence_match(&text, &filter))
}
pub fn fuzzy_like_pattern_with_escape(value: &str, mut escape: impl FnMut(&str) -> String) -> String {
let value = value.trim();
if value.is_empty() {
return "%%".to_string();
}
let mut pattern = String::with_capacity(value.len() * 2 + 2);
pattern.push('%');
for ch in value.chars() {
pattern.push_str(&escape(&ch.to_string()));
pattern.push('%');
}
pattern
}
#[derive(Debug, Clone, Serialize, Deserialize)]
#[serde(rename_all = "camelCase")]
pub struct SqlFileRequest {
pub execution_id: String,
pub connection_id: String,
pub database: String,
/// Optional schema/session namespace for databases where it is distinct
/// from the database used to establish the connection.
#[serde(default, skip_serializing_if = "Option::is_none")]
pub schema: Option<String>,
pub file_path: String,
pub continue_on_error: bool,
/// Reuse a held manual transaction instead of committing through ordinary query execution.
#[serde(default, skip_serializing_if = "Option::is_none")]
pub txn_session_id: Option<String>,
#[serde(default)]
pub selected_tables: Option<Vec<dbx_types::sql_file::SqlFileTable>>,
#[serde(default)]
pub part_cooldown_ms: u64,
/// Temporarily disable relational constraint enforcement across the entire
/// target database for this import, and restore it on completion, error,
/// or cancellation. The mechanism is engine-specific: MySQL-compatible
/// connections toggle session-scoped `FOREIGN_KEY_CHECKS`; PostgreSQL-family
/// connections (Postgres, GaussDB, openGauss) run `ALTER TABLE ... DISABLE
/// TRIGGER ALL` / `ENABLE TRIGGER ALL` for every table; SQL Server runs
/// `ALTER TABLE ... NOCHECK CONSTRAINT ALL` / `WITH NOCHECK CHECK CONSTRAINT
/// ALL` for every table. Has no effect on other connection types. This does
/// not bypass unique/primary-key violations on any engine — only foreign-key
/// and (on PostgreSQL) other trigger-enforced constraints.
#[serde(default)]
pub skip_relational_constraints: bool,
/// Whether the backend writes a JSONL failure log for this execution
/// (dbx#11619). Absent in requests from older frontends, which must keep
/// logging enabled, so the serde default is `true`.
#[serde(default = "sql_file_failure_log_enabled_by_default")]
pub failure_log_enabled: bool,
/// Text encoding used to decode SQL files. `auto` preserves the historical
/// BOM/UTF-8/GBK detection behavior.
#[serde(default)]
pub encoding: SqlFileEncoding,
/// Explicitly allow desktop scripts to include SQL below the entry file's directory.
#[serde(default)]
pub resolve_includes: bool,
}
#[derive(Debug, Clone, Copy, PartialEq, Eq, Serialize, Deserialize, Default)]
#[serde(rename_all = "camelCase")]
pub enum SqlFileEncoding {
#[default]
Auto,
Utf8,
Gbk,
Utf16Le,
Utf16Be,
}
impl SqlFileEncoding {
pub fn to_encoding(self) -> Option<&'static encoding_rs::Encoding> {
match self {
Self::Auto => None,
Self::Utf8 => Some(encoding_rs::UTF_8),
Self::Gbk => Some(encoding_rs::GBK),
Self::Utf16Le => Some(encoding_rs::UTF_16LE),
Self::Utf16Be => Some(encoding_rs::UTF_16BE),
}
}
}
fn sql_file_failure_log_enabled_by_default() -> bool {
true
}
#[derive(Debug, Clone, Serialize, Deserialize)]
#[serde(rename_all = "camelCase")]
pub struct SqlFilePreview {
pub file_name: String,
pub file_path: String,
pub size_bytes: u64,
pub preview: String,
pub can_execute_without_selected_database: bool,
#[serde(default)]
pub establishes_database_context: bool,
#[serde(default, skip_serializing_if = "Option::is_none")]
pub package_file_paths: Option<Vec<String>>,
#[serde(default, skip_serializing_if = "Option::is_none")]
pub package_part_count: Option<usize>,
}
#[derive(Debug, Clone, Copy, PartialEq, Eq, Serialize, Deserialize)]
#[serde(rename_all = "camelCase")]
pub enum SqlFileStatus {
Started,
Running,
StatementDone,
StatementFailed,
Done,
Error,
Cancelled,
}
#[derive(Debug, Clone, PartialEq, Eq)]
pub enum SqlFileStatementAction {
Execute(String),
Skip,
}
#[derive(Debug, Clone, PartialEq, Eq)]
pub enum SqlFileImportStatementKind {
Execute,
Skip,
}
#[derive(Debug, Clone, PartialEq, Eq)]
pub struct SqlFileImportStatement {
pub kind: SqlFileImportStatementKind,
pub sql: String,
pub source_sqls: Vec<String>,
pub source_statement_count: usize,
}
#[derive(Debug, Clone, Copy, PartialEq, Eq)]
struct SqlDialectProfile {
supports_hash_line_comments: bool,
supports_backtick_identifiers: bool,
/// Whether `/* ... */` comments nest. PostgreSQL and SQL Server document
/// nesting (`/* /* */ */` needs both closers, so commenting out a block that
/// already contains a comment keeps the whole block commented); MySQL
/// documents the opposite ("Nested comments are not supported, and are
/// deprecated"). A non-nesting scan ends the comment at the inner `*/` and
/// then hands the commented-out statements to the executor.
supports_nested_block_comments: bool,
/// Whether a backslash inside an ordinary `'...'` string escapes the next
/// character, so `'it\'s'` stays one string.
///
/// This defaults to the historical behaviour (escape) because several
/// engines — MySQL and ClickHouse among them — do escape there, and the ones
/// that do not are listed explicitly below. PostgreSQL and its forks are the
/// engines this repository has evidence for: `standard_conforming_strings` is
/// on by default, which is also why the generated PostgreSQL SQL writes
/// newlines as `E'\n'` instead of `'\n'`.
supports_backslash_escaped_quotes: bool,
/// PostgreSQL escape string literals (`E'...'`), which keep backslash escapes
/// even where ordinary `'...'` literals do not.
supports_postgres_escape_strings: bool,
/// Oracle alternative quoting: `q'[text]'`, `q'{text}'`, `q'(text)'`,
/// `q'<text>'` or `q'XtextX'`. The literal exists so that an apostrophe does
/// not end the string, so reading it as an ordinary `'...'` string ends the
/// text early and splits the statement at a semicolon inside it.
supports_oracle_q_quotes: bool,
supports_oracle_plsql_blocks: bool,
supports_oracle_style_routine_bodies: bool,
supports_slash_line_block_delimiter: bool,
supports_custom_delimiter_commands: bool,
supports_mysql_routine_blocks: bool,
supports_dollar_quoted_strings: bool,
supports_postgres_dollar_quoted_routines: bool,
supports_hana_script_blocks: bool,
supports_firebird_routine_blocks: bool,
supports_go_batch_separator: bool,
keeps_sqlserver_module_batch_at_cursor: bool,
/// Whether a top-level T-SQL control-flow batch (`IF`/`WHILE` with a block
/// body or `ELSE`, or a top-level `BEGIN ... END` block) stays whole in the
/// `;`-statement splitter. The SQL Server profile needs it because T-SQL
/// does not end a block at a semicolon; the generic JDBC profile must not
/// lose it, or the backend re-splits the TRY/CATCH batches the editor
/// already kept together (#13514).
keeps_sqlserver_control_flow_batches: bool,
preserves_tdsql_leading_directives: bool,
requires_whitespace_after_line_comment_dashes: bool,
supports_psql_control_commands: bool,
}
impl Default for SqlDialectProfile {
fn default() -> Self {
Self {
supports_hash_line_comments: false,
supports_backtick_identifiers: true,
supports_nested_block_comments: false,
supports_oracle_q_quotes: false,
supports_backslash_escaped_quotes: true,
supports_postgres_escape_strings: false,
supports_oracle_plsql_blocks: false,
supports_oracle_style_routine_bodies: false,
supports_slash_line_block_delimiter: false,
supports_custom_delimiter_commands: true,
supports_mysql_routine_blocks: false,
supports_dollar_quoted_strings: true,
supports_postgres_dollar_quoted_routines: false,
supports_hana_script_blocks: false,
supports_firebird_routine_blocks: false,
supports_go_batch_separator: false,
keeps_sqlserver_module_batch_at_cursor: false,
keeps_sqlserver_control_flow_batches: false,
preserves_tdsql_leading_directives: false,
requires_whitespace_after_line_comment_dashes: false,
supports_psql_control_commands: false,
}
}
}
impl SqlDialectProfile {
fn for_database_type(db_type: DatabaseType) -> Self {
if matches!(db_type, DatabaseType::Mysql) {
return Self {
preserves_tdsql_leading_directives: crate::tdsql_mysql::preserves_leading_directives_for_database_type(
db_type,
),
..Self::mysql_compatible()
};
}
if matches!(db_type, DatabaseType::Gaussdb) {
return Self::gaussdb();
}
if matches!(db_type, DatabaseType::Postgres | DatabaseType::OpenGauss) {
return Self::postgres_family();
}
if matches!(db_type, DatabaseType::OceanbaseOracle) {
return Self {
supports_backtick_identifiers: false,
supports_dollar_quoted_strings: false,
supports_custom_delimiter_commands: false,
..Self::oracle_like()
};
}
if Self::is_oracle_like_database(db_type) {
return Self::oracle_like();
}
if matches!(db_type, DatabaseType::SqlServer) {
return Self::sql_server();
}
if matches!(db_type, DatabaseType::Db2) {
// DB2 block comments nest; cursor extraction must skip the complete
// leading comment before handing the query to the Explain gate.
return Self { supports_nested_block_comments: true, ..Self::default() };
}
if matches!(db_type, DatabaseType::ClickHouse) {
// ClickHouse documents that C-style block comments nest.
return Self { supports_nested_block_comments: true, ..Self::default() };
}
if Self::is_mysql_compatible_database(db_type) {
return Self::mysql_compatible();
}
if matches!(db_type, DatabaseType::SapHana) {
return Self::sap_hana();
}
if matches!(db_type, DatabaseType::Firebird) {
return Self::firebird();
}
if matches!(db_type, DatabaseType::DuckDb) {
return Self {
supports_backslash_escaped_quotes: false,
supports_postgres_escape_strings: true,
..Self::default()
};
}
if Self::is_postgres_string_lexer_database(db_type) {
return Self {
supports_backslash_escaped_quotes: false,
supports_nested_block_comments: true,
supports_postgres_escape_strings: true,
..Self::default()
};
}
if matches!(db_type, DatabaseType::Jdbc | DatabaseType::Hive) {
return Self { supports_oracle_style_routine_bodies: true, ..Self::default() };
}
Self::default()
}
/// PostgreSQL and the forks that inherit its string literal rules without
/// inheriting the routine/delimiter handling of [`Self::postgres_family`].
///
/// Redshift is deliberately absent: it descends from PostgreSQL 8.0, which
/// predates `standard_conforming_strings` defaulting to `on`, so it keeps the
/// historical behaviour until its own evidence shows otherwise.
fn is_postgres_string_lexer_database(db_type: DatabaseType) -> bool {
matches!(
db_type,
DatabaseType::Vastbase
| DatabaseType::Kingbase
| DatabaseType::Highgo
| DatabaseType::Uxdb
| DatabaseType::Gbase8c
| DatabaseType::Kwdb
)
}
fn mysql_compatible() -> Self {
Self {
supports_hash_line_comments: true,
supports_backslash_escaped_quotes: true,
supports_mysql_routine_blocks: true,
requires_whitespace_after_line_comment_dashes: true,
..Self::default()
}
}
fn oracle_like() -> Self {
Self {
supports_backslash_escaped_quotes: false,
supports_oracle_plsql_blocks: true,
supports_oracle_q_quotes: true,
supports_slash_line_block_delimiter: true,
..Self::default()
}
}
/// PostgreSQL and openGauss, which are also the connection types users pick for a
/// GaussDB/openGauss instance running in Oracle (A) compatibility mode. GaussDB PL/SQL
/// bodies are Oracle-style (`CREATE PROCEDURE p AS DECLARE ... BEGIN ... END;`) and are
/// terminated by a standalone `/` line, while the surrounding script syntax is still
/// PostgreSQL, so the profile keeps those constructs whole without enabling the full
/// Oracle PL/SQL dialect.
fn postgres_family() -> Self {
Self {
supports_backslash_escaped_quotes: false,
supports_nested_block_comments: true,
supports_oracle_style_routine_bodies: true,
supports_postgres_escape_strings: true,
supports_slash_line_block_delimiter: true,
supports_psql_control_commands: true,
..Self::default()
}
}
fn gaussdb() -> Self {
Self {
supports_backslash_escaped_quotes: false,
supports_nested_block_comments: true,
supports_postgres_dollar_quoted_routines: true,
supports_postgres_escape_strings: true,
supports_psql_control_commands: true,
..Self::oracle_like()
}
}
fn sql_server() -> Self {
Self {
supports_go_batch_separator: true,
keeps_sqlserver_module_batch_at_cursor: true,
keeps_sqlserver_control_flow_batches: true,
supports_nested_block_comments: true,
..Self::default()
}
}
fn sap_hana() -> Self {
Self { supports_hana_script_blocks: true, ..Self::default() }
}
fn firebird() -> Self {
Self { supports_backslash_escaped_quotes: false, supports_firebird_routine_blocks: true, ..Self::default() }
}
fn is_mysql_compatible_database(db_type: DatabaseType) -> bool {
matches!(
db_type,
DatabaseType::Mysql
| DatabaseType::Doris
| DatabaseType::StarRocks
| DatabaseType::ManticoreSearch
| DatabaseType::Goldendb
)
}
fn is_oracle_like_database(db_type: DatabaseType) -> bool {
matches!(
db_type,
DatabaseType::Oracle
| DatabaseType::Dameng
| DatabaseType::Gaussdb
| DatabaseType::Yashandb
| DatabaseType::Oscar
| DatabaseType::OceanbaseOracle
| DatabaseType::Xugu
// ArgoDB (Transwarp fork of Hive/Inceptor) ships a PL/SQL-compatible procedure
// language: `CREATE [OR REPLACE] PROCEDURE ... IS BEGIN ... END;`. Treat it like
// Oracle for statement splitting so semicolons inside the procedure body are not
// misinterpreted as client-side statement terminators.
| DatabaseType::Argo
| DatabaseType::Transwarp
)
}
}
#[derive(Debug, Clone, Copy, Default)]
pub struct SqlParsingOptions {
profile: SqlDialectProfile,
}
impl SqlParsingOptions {
pub fn for_database_type(db_type: DatabaseType) -> Self {
Self::from_profile(SqlDialectProfile::for_database_type(db_type))
}
pub fn for_database_type_and_compatibility(db_type: DatabaseType, compatibility_mode: Option<&str>) -> Self {
if db_type == DatabaseType::Kingbase {
let mode = compatibility_mode.unwrap_or_default().trim().to_ascii_lowercase();
return Self::from_profile(match mode.as_str() {
"mysql" => SqlDialectProfile::mysql_compatible(),
"sqlserver" => SqlDialectProfile::sql_server(),
"pg" | "postgres" | "postgresql" => SqlDialectProfile::postgres_family(),
// Preserve PL/SQL blocks while mode detection is unavailable;
// the GaussDB profile also preserves dollar-quoted routines.
_ => SqlDialectProfile::gaussdb(),
});
}
if db_type == DatabaseType::OpenGauss {
return match compatibility_mode.map(str::trim) {
// openGauss stores package specs/bodies as PL/SQL regardless of
// mode, and A mode is the only mode where the catalog reports
// them as packages. Any mode other than A keeps the PostgreSQL
// statement splitter.
Some(mode) if mode.eq_ignore_ascii_case("A") => Self::from_profile(SqlDialectProfile::gaussdb()),
Some(_) => Self::for_database_type(db_type),
// Unknown mode: the compatibility probe failed or the pool was
// unavailable. Falling back to the plain PostgreSQL profile would
// split an A-mode package body on its inner semicolons into
// fragments that the caller may then execute individually, so the
// conservative PL/SQL-capable profile is used instead.
None => Self::from_profile(SqlDialectProfile::gaussdb()),
};
}
Self::for_database_type(db_type)
}
pub fn mysql_compatible() -> Self {
Self::from_profile(SqlDialectProfile::mysql_compatible())
}
fn from_profile(profile: SqlDialectProfile) -> Self {
Self { profile }
}
}
#[derive(Debug, Clone, Serialize, Deserialize)]
#[serde(rename_all = "camelCase")]
pub struct SqlFileProgress {
pub execution_id: String,
pub status: SqlFileStatus,
pub statement_index: usize,
pub success_count: usize,
pub failure_count: usize,
pub affected_rows: u64,
pub elapsed_ms: u128,
pub statement_summary: String,
pub error: Option<String>,
#[serde(default, skip_serializing_if = "Option::is_none")]
pub bytes_read: Option<u64>,
#[serde(default, skip_serializing_if = "Option::is_none")]
pub total_bytes: Option<u64>,
#[serde(default, skip_serializing_if = "Option::is_none")]
pub phase: Option<SqlFilePhase>,
/// When processing multiple files, the 0-based index of the current file.
#[serde(skip_serializing_if = "Option::is_none")]
pub file_index: Option<usize>,
/// When processing multiple files, the name of the current file.
#[serde(skip_serializing_if = "Option::is_none")]
pub file_name: Option<String>,
/// Full source SQL behind a `StatementFailed` event, used by the backend to
/// write the failure log. It is deliberately never serialized: the frontend
/// only ever needs the 120-char summary, and shipping every failing statement
/// over IPC would bloat the progress stream it already throttles.
#[serde(skip)]
#[serde(default)]
pub failure_sql: Option<String>,
}
#[derive(Debug, Clone, Copy, PartialEq, Eq, Serialize, Deserialize)]
#[serde(rename_all = "camelCase")]
pub enum SqlFilePhase {
Preparing,
Reading,
Executing,
}
pub fn decode_sql_file_bytes(bytes: &[u8]) -> Result<String, String> {
if bytes.starts_with(&[0xEF, 0xBB, 0xBF]) {
return std::str::from_utf8(&bytes[3..]).map(|text| text.to_string()).map_err(|_| sql_file_encoding_error());
}
if bytes.starts_with(&[0xFF, 0xFE]) {
return decode_sql_file_with_encoding(&bytes[2..], encoding_rs::UTF_16LE);
}
if bytes.starts_with(&[0xFE, 0xFF]) {
return decode_sql_file_with_encoding(&bytes[2..], encoding_rs::UTF_16BE);
}
if let Ok(text) = std::str::from_utf8(bytes) {
return Ok(text.strip_prefix('\u{feff}').unwrap_or(text).to_string());
}
decode_sql_file_with_encoding(bytes, encoding_rs::GBK)
}
/// Decode an SQL/text file using an explicit user-selected encoding.
pub fn decode_sql_file_bytes_with_encoding(bytes: &[u8], encoding: &str) -> Result<String, String> {
match encoding {
"utf8" => {
decode_sql_file_with_encoding(bytes.strip_prefix(&[0xEF, 0xBB, 0xBF]).unwrap_or(bytes), encoding_rs::UTF_8)
}
"utf8Bom" => {
let payload = bytes.strip_prefix(&[0xEF, 0xBB, 0xBF]).unwrap_or(bytes);
decode_sql_file_with_encoding(payload, encoding_rs::UTF_8)
}
"utf16le" => {
decode_sql_file_with_encoding(bytes.strip_prefix(&[0xFF, 0xFE]).unwrap_or(bytes), encoding_rs::UTF_16LE)
}
"utf16be" => {
decode_sql_file_with_encoding(bytes.strip_prefix(&[0xFE, 0xFF]).unwrap_or(bytes), encoding_rs::UTF_16BE)
}
"gbk" => decode_sql_file_with_encoding(bytes, encoding_rs::GBK),
"auto" | "" => decode_sql_file_bytes(bytes),
_ => Err(format!("Unsupported SQL file encoding: {encoding}")),
}
}
/// Encode editor text for an explicit file encoding. The returned bytes include
/// a BOM for UTF-8 BOM and UTF-16 variants.
pub fn encode_sql_file_text(content: &str, encoding: &str) -> Result<Vec<u8>, String> {
if encoding == "utf16le" || encoding == "utf16be" {
let mut bytes = if encoding == "utf16le" { vec![0xFF, 0xFE] } else { vec![0xFE, 0xFF] };
for unit in content.encode_utf16() {
bytes.extend_from_slice(&if encoding == "utf16le" { unit.to_le_bytes() } else { unit.to_be_bytes() });
}
return Ok(bytes);
}
let (encoded, _, had_errors) = match encoding {
"utf8" | "auto" | "" => encoding_rs::UTF_8.encode(content),
"utf8Bom" => encoding_rs::UTF_8.encode(content),
"gbk" => encoding_rs::GBK.encode(content),
_ => return Err(format!("Unsupported SQL file encoding: {encoding}")),
};
if had_errors {
return Err(sql_file_encoding_error());
}
let mut bytes = encoded.into_owned();
if encoding == "utf8Bom" {
bytes.splice(0..0, [0xEF, 0xBB, 0xBF]);
}
Ok(bytes)
}
fn decode_sql_file_with_encoding(bytes: &[u8], encoding: &'static encoding_rs::Encoding) -> Result<String, String> {
let (text, had_errors) = encoding.decode_without_bom_handling(bytes);
if had_errors {
return Err(sql_file_encoding_error());
}
Ok(text.into_owned())
}
fn sql_file_encoding_error() -> String {
"Unsupported SQL file encoding. Save the file as UTF-8, UTF-8 with BOM, UTF-16 with BOM, or GBK, then try again."
.to_string()
}
/// Whether a `--` sequence starts a line comment. MySQL (and MySQL-wire-compatible
/// dialects) requires the second dash to be followed by whitespace, a control character,
/// or end-of-input —
/// `5--1` is subtraction (`5 - -1`), not a comment. Other dialects (Postgres, Oracle/DM,
/// SQL Server, ...) treat `--` as a comment opener unconditionally.
fn dash_dash_starts_line_comment(profile: SqlDialectProfile, char_after_dashes: Option<char>) -> bool {
if !profile.requires_whitespace_after_line_comment_dashes {
return true;
}
char_after_dashes.is_none_or(|ch| ch.is_whitespace() || ch.is_control())
}
/// Whether the text ends with an XML/HTML character reference whose closing `;`
/// is about to be scanned. MyBatis XML writes comparison operators as `>=`
/// / `<=`, and that `;` must not be read as a statement terminator (#11699).
/// Only the five predefined XML names (`lt`, `gt`, `amp`, `quot`, `apos`) plus
/// numeric references (`'`, `'`) are recognized, so a plain bitwise
/// `a&b;` still splits at its real terminator. Mirrors
/// `isXmlEntityReferenceSemicolon` in the editor splitter
/// (`apps/desktop/src/lib/sql/sqlStatementRanges.ts`); keep both in sync.
fn ends_with_xml_entity_reference(buffer: &str) -> bool {
let bytes = buffer.as_bytes();
let mut start = bytes.len();
while start > 0 && bytes[start - 1].is_ascii_alphanumeric() {
start -= 1;
}
if start > 0 && bytes[start - 1] == b'#' {
start -= 1;
}
if start == 0 || bytes[start - 1] != b'&' {
return false;
}
let name = &buffer[start..];
if let Some(hex) = name.strip_prefix("#x").or_else(|| name.strip_prefix("#X")) {
return !hex.is_empty() && hex.bytes().all(|byte| byte.is_ascii_hexdigit());
}
if let Some(digits) = name.strip_prefix('#') {
return !digits.is_empty() && digits.bytes().all(|byte| byte.is_ascii_digit());
}
matches!(name, "lt" | "gt" | "amp" | "quot" | "apos")
}
/// Tracks an Oracle `q'...'` or `nq'...'` literal across characters, so a semicolon inside its
/// text never splits the statement.
#[derive(Default, Clone, Copy)]
struct OracleQQuoteState {
awaiting_delimiter: bool,
closer: Option<char>,
}
impl OracleQQuoteState {
fn is_open(&self) -> bool {
self.awaiting_delimiter || self.closer.is_some()
}
/// Consumes one character of an open literal; `previous` is the character
/// before it, as tracked by the caller.
fn consume(&mut self, ch: char, previous: Option<char>) {
if self.awaiting_delimiter {
self.closer = oracle_q_quote_closer(ch);
self.awaiting_delimiter = false;
} else if let Some(closer) = self.closer {
if previous == Some(closer) && ch == '\'' {
self.closer = None;
}
}
}
}
#[derive(Default)]
pub struct SqlStatementSplitter {
buffer: String,
in_single_quote: bool,
/// Whether the open `'...'` literal reads a backslash as escaping the next
/// character: MySQL strings always do, PostgreSQL only inside `E'...'`.
single_quote_escape_string: bool,
in_double_quote: bool,
in_backtick: bool,
in_line_comment: bool,
/// Depth of the open `/* ... */` comment. Engines that nest block comments
/// increment it for every inner `/*`, so the inner `*/` does not end the
/// comment and hand commented-out statements to the executor.
block_comment_depth: usize,
dollar_quote_tag: Option<String>,
postgres_dollar_quoted_routine: bool,
previous: Option<char>,
pending_mysql_line_comment_dashes: bool,
/// Open Oracle `q'...'` literal, if any.
oracle_q_quote: OracleQQuoteState,
/// PL/SQL block state of the buffered statement, advanced at each `;`.
oracle_block: OraclePlSqlBlockTracker,
custom_delimiter: Option<String>,
stop_on_error: bool,
pending_psql_command: Option<String>,
options: SqlParsingOptions,
}
#[derive(Debug, Clone, PartialEq, Eq)]
pub struct SqlStatementWithControl {
pub sql: String,
pub stop_on_error: bool,
}
impl SqlStatementSplitter {
pub fn with_options(options: SqlParsingOptions) -> Self {
Self { options, ..Self::default() }
}
/// A whole-line file directive must not be interpreted inside SQL text.
pub fn can_start_file_directive(&self) -> bool {
!self.in_single_quote
&& !self.in_double_quote
&& !self.in_backtick
&& !self.in_line_comment
&& self.block_comment_depth == 0
&& self.dollar_quote_tag.is_none()
&& !self.oracle_q_quote.is_open()
&& self.pending_psql_command.is_none()
}
pub fn is_at_file_directive_boundary(&self) -> bool {
self.can_start_file_directive() && !has_executable_sql_with_options(&self.buffer, self.options)
}
pub fn push_chunk(&mut self, chunk: &str) -> Vec<String> {
self.push_chunk_with_control(chunk).into_iter().map(|statement| statement.sql).collect()
}
pub fn push_chunk_with_control(&mut self, chunk: &str) -> Vec<SqlStatementWithControl> {
let mut statements = Vec::new();
let chars = chunk.chars().collect::<Vec<_>>();
let mut i = 0;
if self.pending_mysql_line_comment_dashes {
if let Some(first) = chars.first().copied() {
self.in_line_comment = dash_dash_starts_line_comment(self.options.profile, Some(first));
self.pending_mysql_line_comment_dashes = false;
}
}
while i < chars.len() {
if let Some(command) = &mut self.pending_psql_command {
let ch = chars[i];
command.push(ch);
if ch == '\n' {
self.finish_psql_command(&mut statements);
}
i += 1;
continue;
}
// A psql command consumes the whole line, including quotes, comment
// markers and semicolons in display text. Buffer it across chunks
// before letting the SQL scanner interpret those characters.
if self.options.profile.supports_psql_control_commands
&& chars[i] == '\\'
&& !self.in_single_quote
&& !self.in_double_quote
&& !self.in_backtick
&& !self.in_line_comment
&& self.block_comment_depth == 0
&& self.dollar_quote_tag.is_none()
&& self.buffer.rsplit('\n').next().unwrap_or("").trim().is_empty()
{
self.pending_psql_command = Some(String::from("\\"));
i += 1;
continue;
}
if let Some(tag) = &self.dollar_quote_tag {
let tag_chars = tag.chars().collect::<Vec<_>>();
if starts_with_chars(&chars, i, &tag_chars) {
for tag_ch in &tag_chars {
self.buffer.push(*tag_ch);
self.previous = Some(*tag_ch);
}
i += tag_chars.len();
self.dollar_quote_tag = None;
continue;
}
let ch = chars[i];
self.buffer.push(ch);
self.previous = Some(ch);
i += 1;
continue;
}
let ch = chars[i];
let next = chars.get(i + 1).copied();
if self.in_line_comment {
self.buffer.push(ch);
if ch == '\n' {
self.in_line_comment = false;
}
self.previous = Some(ch);
i += 1;
continue;
}
if self.block_comment_depth > 0 {
self.buffer.push(ch);
if self.options.profile.supports_nested_block_comments && self.previous == Some('/') && ch == '*' {
self.block_comment_depth += 1;
// A nested opener's asterisk must not pair with a following slash as a close.
self.previous = None;
i += 1;
continue;
} else if self.previous == Some('*') && ch == '/' {
self.block_comment_depth -= 1;
}
self.previous = Some(ch);
i += 1;
continue;
}
if self.oracle_q_quote.is_open() {
self.oracle_q_quote.consume(ch, self.previous);
self.buffer.push(ch);
self.previous = Some(ch);
i += 1;
continue;
}
if !self.in_single_quote && !self.in_double_quote && !self.in_backtick {
if self.options.profile.supports_oracle_q_quotes
&& ch == '\''
&& ends_with_oracle_q_quote_prefix(&self.buffer)
{
self.oracle_q_quote.awaiting_delimiter = true;
self.buffer.push(ch);
self.previous = Some(ch);
i += 1;
continue;
}
if self.previous == Some('-') && ch == '-' {
if self.options.profile.requires_whitespace_after_line_comment_dashes && next.is_none() {
self.pending_mysql_line_comment_dashes = true;
self.buffer.push(ch);
self.previous = Some(ch);
i += 1;
continue;
}
if dash_dash_starts_line_comment(self.options.profile, next) {
self.in_line_comment = true;
self.buffer.push(ch);
self.previous = Some(ch);
i += 1;
continue;
}
}
if self.previous == Some('/') && ch == '*' {
self.block_comment_depth = 1;
self.buffer.push(ch);
// The opener's own asterisk must not read as a nested opener.
self.previous = None;
i += 1;
continue;
}
if self.options.profile.supports_hash_line_comments && ch == '#' {
self.in_line_comment = true;
self.buffer.push(ch);
self.previous = Some(ch);
i += 1;
continue;
}
if ch == '/' && next == Some('*') {
self.block_comment_depth = 1;
self.buffer.push(ch);
// The opener's own asterisk must not read as a nested opener.
self.previous = None;
i += 1;
continue;
}
if let Some(tag) = self
.options
.profile
.supports_dollar_quoted_strings
.then(|| dollar_quote_tag_at(&chars, i))
.flatten()
{
if self.custom_delimiter.is_none() && !self.on_delimiter_line() {
if self.options.profile.supports_postgres_dollar_quoted_routines
&& starts_with_postgres_dollar_quoted_routine_prefix(&self.buffer)
{
self.postgres_dollar_quoted_routine = true;
}
for tag_ch in tag.chars() {
self.buffer.push(tag_ch);
self.previous = Some(tag_ch);
}
i += tag.chars().count();
self.dollar_quote_tag = Some(tag);
continue;
}
}
}
match ch {
'\'' if !(self.in_double_quote
|| self.in_backtick
|| self.single_quote_backslash_escapes() && has_odd_trailing_backslashes(&self.buffer)) =>
{
self.in_single_quote = !self.in_single_quote;
self.single_quote_escape_string = self.in_single_quote
&& (self.options.profile.supports_backslash_escaped_quotes
|| (self.options.profile.supports_postgres_escape_strings
&& ends_with_escape_string_prefix(&self.buffer)));
self.buffer.push(ch);
}
'"' if !(self.in_single_quote
|| self.in_backtick
|| self.options.profile.supports_backslash_escaped_quotes
&& has_odd_trailing_backslashes(&self.buffer)) =>
{
self.in_double_quote = !self.in_double_quote;
self.buffer.push(ch);
}
'`' if self.options.profile.supports_backtick_identifiers
&& !self.in_single_quote
&& !self.in_double_quote =>
{
self.in_backtick = !self.in_backtick;
self.buffer.push(ch);
}
';' if !self.in_single_quote && !self.in_double_quote && !self.in_backtick => {
if ends_with_xml_entity_reference(&self.buffer) {
// The `;` closes an XML/MyBatis entity (`>`, `<`,
// `'`, ...), not the statement (#11699).
self.buffer.push(ch);
} else if (self.options.profile.supports_custom_delimiter_commands && self.on_delimiter_line())
|| self.custom_delimiter.is_some()
{
self.buffer.push(ch);
} else if self.options.profile.supports_mysql_routine_blocks
&& starts_with_mysql_routine_block(&self.buffer)
{
let mut candidate = self.buffer.clone();
candidate.push(ch);
if mysql_routine_block_is_complete(&candidate) {
// The final semicolon is the client-side statement delimiter.
// Keep semicolons inside BEGIN...END, but do not send the
// delimiter after END to the MySQL server.
self.push_current_statement(&mut statements);
} else {
self.buffer.push(ch);
}
} else if !self.postgres_dollar_quoted_routine
&& self.oracle_block.keeps_block_together(self.options, &self.buffer)
{
self.buffer.push(ch);
if self.oracle_block.is_complete(&self.buffer) {
self.push_current_statement(&mut statements);
}
} else if self.options.profile.supports_hana_script_blocks
&& starts_with_hana_script_block(&self.buffer)
{
self.buffer.push(ch);
if hana_script_block_is_complete(&self.buffer) {
self.push_current_statement(&mut statements);
}
} else if self.options.profile.supports_firebird_routine_blocks
&& starts_with_firebird_routine_block(&self.buffer)
{
self.buffer.push(ch);
if firebird_routine_block_is_complete(&self.buffer) {
self.push_current_statement(&mut statements);
}
} else if keeps_sqlserver_control_flow_batch(self.options, &self.buffer) {
// T-SQL does not end a control-flow batch at a semicolon
// (#9336, #13051, #13514): the inner `;` stays in the
// batch and the statement ends after the block's final
// `END [TRY]`.
self.buffer.push(ch);
if sqlserver_control_flow_batch_is_complete(&self.buffer) {
// A final `END` without its own `;` shares the
// fragment with the next statement; the boundary
// sits right after that `END` instead (TS parity).
// An `END` directly followed by this `;` keeps the
// historical shape: the batch ends here with its
// terminating semicolon.
let body_end = self.buffer.len() - ch.len_utf8();
let body = self.buffer[..body_end].to_string();
let mut boundaries = Vec::new();
let mut cursor = 0usize;
while let Some(relative) = sqlserver_control_flow_batch_boundary(&body[cursor..]) {
if cursor + relative >= body.len() {
break;
}
cursor += relative;
boundaries.push(cursor);
}
if boundaries.is_empty() {
self.push_current_statement(&mut statements);
} else {
let mut piece_from = 0usize;
for boundary_end in &boundaries {
self.buffer.clear();
self.buffer.push_str(&body[piece_from..*boundary_end]);
self.push_current_statement(&mut statements);
piece_from = *boundary_end;
}
let trailing = &body[piece_from..];
if !trailing.trim().is_empty() {
self.buffer.clear();
self.buffer.push_str(trailing);
if keeps_sqlserver_control_flow_batch(self.options, trailing) {
// A new batch in the remainder keeps the
// terminating `;` when complete and this
// `;` when it is still mid-typing.
self.buffer.push(ch);
if sqlserver_control_flow_batch_is_complete(trailing) {
self.push_current_statement(&mut statements);
}
} else {
self.push_current_statement(&mut statements);
}
} else {
self.buffer.clear();
}
}
}
} else {
self.push_current_statement(&mut statements);
}
}
_ => self.buffer.push(ch),
}
if !self.in_single_quote && !self.in_double_quote && !self.in_backtick && self.dollar_quote_tag.is_none() {
if ch == '\n' {
let buf_end = self.buffer.len() - 1;
let last_line_start = self.buffer[..buf_end].rfind('\n').map_or(0, |p| p + 1);
let last_line = self.buffer[last_line_start..buf_end].trim();
if self.options.profile.supports_slash_line_block_delimiter && last_line == "/" {
let before = self.buffer[..last_line_start].trim();
if has_executable_sql_with_options(before, self.options) {
statements.push(SqlStatementWithControl {
sql: before.to_string(),
stop_on_error: self.stop_on_error,
});
}
self.buffer.clear();
self.oracle_block.reset();
self.postgres_dollar_quoted_routine = false;
self.previous = None;
i += 1;
continue;
}
if let Some(new_delim) = self
.options
.profile
.supports_custom_delimiter_commands
.then(|| {
parse_delimiter_command(last_line).filter(|_| {
!has_executable_sql_with_options(&self.buffer[..last_line_start], self.options)
})
})
.flatten()
{
self.custom_delimiter = if new_delim == ";" { None } else { Some(new_delim.to_string()) };
if last_line_start > 0 {
let before = self.buffer[..last_line_start].trim();
if has_executable_sql_with_options(before, self.options) {
statements.push(SqlStatementWithControl {
sql: before.to_string(),
stop_on_error: self.stop_on_error,
});
}
}
self.buffer.clear();
self.oracle_block.reset();
self.postgres_dollar_quoted_routine = false;
self.previous = None;
i += 1;
continue;
}
}
if let Some(delim) = self.custom_delimiter.clone() {
// A DELIMITER command can itself end with the active
// delimiter (for example `DELIMITER ;;`). Keep the
// directive intact so the newline handler can switch the
// delimiter instead of emitting a bogus `delimiter`
// statement or advancing past the line start.
if self.buffer.ends_with(delim.as_str()) && !self.on_delimiter_line() {
self.buffer.truncate(self.buffer.len() - delim.len());
self.push_current_statement(&mut statements);
}
}
}
self.previous = Some(ch);
i += 1;
}
statements
}
fn finish_psql_command(&mut self, statements: &mut Vec<SqlStatementWithControl>) {
let command = self.pending_psql_command.take().expect("pending psql command");
if let Some(control) = parse_ignorable_psql_control_command(command.trim()) {
self.stop_on_error |= control == PsqlControlCommand::OnErrorStop;
} else {
// Unsupported commands must reach the SQL executor and fail visibly;
// do not silently discard variable assignments or execution commands.
self.options.profile.supports_psql_control_commands = false;
statements.extend(self.push_chunk_with_control(&command));
self.options.profile.supports_psql_control_commands = true;
}
}
pub fn finish(self) -> Vec<String> {
self.finish_with_control().into_iter().map(|statement| statement.sql).collect()
}
pub fn finish_with_control(mut self) -> Vec<SqlStatementWithControl> {
let mut statements = Vec::new();
if self.pending_psql_command.is_some() {
self.finish_psql_command(&mut statements);
}
if self.pending_mysql_line_comment_dashes {
self.in_line_comment = true;
self.pending_mysql_line_comment_dashes = false;
}
let trimmed = self.buffer.trim();
let last_line = trimmed.rsplit('\n').next().unwrap_or(trimmed).trim();
let before_last_line = trimmed.rsplit_once('\n').map(|x| x.0).unwrap_or("").trim();
if self.options.profile.supports_custom_delimiter_commands
&& parse_delimiter_command(last_line)
.is_some_and(|_| !has_executable_sql_with_options(before_last_line, self.options))
{
let before = trimmed.rsplit_once('\n').map(|x| x.0).unwrap_or("").trim();
if has_executable_sql_with_options(before, self.options) {
statements.push(SqlStatementWithControl { sql: before.to_string(), stop_on_error: self.stop_on_error });
}
self.buffer.clear();
} else if self.options.profile.supports_slash_line_block_delimiter && last_line == "/" {
let before = trimmed.rsplit_once('\n').map(|x| x.0).unwrap_or("").trim();
if has_executable_sql_with_options(before, self.options) {
statements.push(SqlStatementWithControl { sql: before.to_string(), stop_on_error: self.stop_on_error });
}
self.buffer.clear();
} else if let Some(ref delim) = self.custom_delimiter {
if self.buffer.ends_with(delim.as_str()) {
self.buffer.truncate(self.buffer.len() - delim.len());
}
}
self.push_current_statement(&mut statements);
statements
}
pub fn stop_on_error(&self) -> bool {
self.stop_on_error
}
fn push_current_statement(&mut self, statements: &mut Vec<SqlStatementWithControl>) {
let statement = self.buffer.trim();
if has_executable_sql_with_options(statement, self.options) {
statements.push(SqlStatementWithControl { sql: statement.to_string(), stop_on_error: self.stop_on_error });
}
self.buffer.clear();
self.oracle_block.reset();
self.postgres_dollar_quoted_routine = false;
self.previous = None;
self.pending_mysql_line_comment_dashes = false;
}
/// A backslash escapes a quote only where the dialect says so: MySQL for
/// ordinary strings, PostgreSQL for `E'...'` literals. Outside a string the
/// profile decides, which keeps the historical MySQL guard that a quote
/// preceded by an odd number of backslashes does not open a literal.
fn single_quote_backslash_escapes(&self) -> bool {
if self.in_single_quote {
self.single_quote_escape_string
} else {
self.options.profile.supports_backslash_escaped_quotes
}
}
fn on_delimiter_line(&self) -> bool {
let start = self.buffer.rfind('\n').map_or(0, |p| p + 1);
let line = self.buffer[start..].trim_start().as_bytes();
line.len() >= 9
&& line[..9].eq_ignore_ascii_case(b"delimiter")
&& !has_executable_sql_with_options(self.buffer[..start].trim(), self.options)
}
}
fn has_odd_trailing_backslashes(sql: &str) -> bool {
sql.as_bytes().iter().rev().take_while(|byte| **byte == b'\\').count() % 2 == 1
}
/// True when `text` ends with a PostgreSQL `E'`/`e'` escape-string introducer.
/// The `E` must start its own token: a type name that merely ends in `e`
/// (`DATE'2020-01-01'`) or an identifier such as `x$e` is not an escape string.
fn ends_with_escape_string_prefix(text: &str) -> bool {
let mut chars = text.chars().rev();
if !matches!(chars.next(), Some('E' | 'e')) {
return false;
}
!chars.next().is_some_and(is_identifier_continue_char)
}
/// Whether `ch` can continue an unquoted identifier, which decides whether a
/// one-letter introducer (`E'...'`, `q'...'`) starts its own token.
fn is_identifier_continue_char(ch: char) -> bool {
ch.is_alphanumeric() || ch == '_' || ch == '$'
}
/// Whether the text before a quote ends in an Oracle `q`/`nq` introducer token.
pub(crate) fn ends_with_oracle_q_quote_prefix(text: &str) -> bool {
let mut chars = text.chars().rev();
if !matches!(chars.next(), Some('q' | 'Q')) {
return false;
}
let mut previous = chars.next();
if matches!(previous, Some('n' | 'N')) {
previous = chars.next();
}
!previous.is_some_and(is_identifier_continue_char)
}
/// The closing character of an Oracle alternative-quoting delimiter: brackets
/// pair up (`q'[a]'`) and every other character delimits itself (`q'!a!'`).
/// A quote or whitespace cannot delimit the literal.
fn oracle_q_quote_closer(opening: char) -> Option<char> {
match opening {
'[' => Some(']'),
'{' => Some('}'),
'(' => Some(')'),
'<' => Some('>'),
'\'' | ' ' | '\t' | '\n' | '\r' => None,
other => Some(other),
}
}
/// Byte index just past the closing quote of the `q'...'` literal whose opening
/// quote sits at `quote_index`, or `None` when the literal is not well formed.
pub(crate) fn oracle_q_quote_end(sql: &str, quote_index: usize) -> Option<usize> {
let opening = next_char_at(sql, quote_index + 1)?;
let closer = oracle_q_quote_closer(opening)?;
let content_start = quote_index + 1 + opening.len_utf8();
let closing = format!("{closer}'");
let offset = sql.get(content_start..)?.find(&closing)?;
Some(content_start + offset + closing.len())
}
pub fn split_sql_statements(sql: &str) -> Vec<String> {
split_sql_statements_with_options(sql, SqlParsingOptions::default())
}
#[cfg(test)]
#[test]
fn kingbase_compatibility_profiles_preserve_routine_and_batch_boundaries() {
for (mode, db_type, sql) in [
("oracle", DatabaseType::Gaussdb, "BEGIN\n NULL;\n NULL;\nEND;\n/\nSELECT 1;"),
(
"mysql",
DatabaseType::Mysql,
"DELIMITER $$\nCREATE PROCEDURE p() BEGIN SELECT 1; SELECT 2; END$$\nDELIMITER ;\nSELECT 3;",
),
(
"pg",
DatabaseType::Postgres,
"CREATE FUNCTION f() RETURNS int AS $$ BEGIN RETURN 1; END; $$ LANGUAGE plpgsql; SELECT 2;",
),
(
"sqlserver",
DatabaseType::SqlServer,
"CREATE PROCEDURE p AS\nBEGIN\n SELECT 1;\n SELECT 2;\nEND\nGO\nSELECT 3;",
),
] {
assert_eq!(
split_sql_statements_for_database_with_compatibility(sql, DatabaseType::Kingbase, Some(mode)),
split_sql_statements_for_database(sql, db_type),
"Kingbase mode {mode}"
);
}
}
pub fn split_sql_statements_for_database(sql: &str, db_type: DatabaseType) -> Vec<String> {
sql_execution_plan_for_database(sql, db_type).statements
}
pub fn split_sql_statements_for_database_with_compatibility(
sql: &str,
db_type: DatabaseType,
compatibility_mode: Option<&str>,
) -> Vec<String> {
sql_execution_plan_for_database_with_compatibility(sql, db_type, compatibility_mode).statements
}
#[derive(Debug, Clone, PartialEq, Eq)]
pub struct SqlExecutionPlan {
pub statements: Vec<String>,
pub stop_on_error: bool,
}
pub fn sql_execution_plan_for_database(sql: &str, db_type: DatabaseType) -> SqlExecutionPlan {
sql_execution_plan_with_options(sql, SqlParsingOptions::for_database_type(db_type))
}
pub fn sql_execution_plan_for_database_with_compatibility(
sql: &str,
db_type: DatabaseType,
compatibility_mode: Option<&str>,
) -> SqlExecutionPlan {
sql_execution_plan_with_options(
sql,
SqlParsingOptions::for_database_type_and_compatibility(db_type, compatibility_mode),
)
}
fn sql_execution_plan_with_options(sql: &str, options: SqlParsingOptions) -> SqlExecutionPlan {
if !options.profile.supports_psql_control_commands {
return SqlExecutionPlan { statements: split_sql_statements_with_options(sql, options), stop_on_error: false };
}
let (sql, stop_on_error) = preprocess_psql_control_commands(sql, options.profile);
let mut parsing_options = options;
parsing_options.profile.supports_psql_control_commands = false;
SqlExecutionPlan { statements: split_sql_statements_with_options(&sql, parsing_options), stop_on_error }
}
pub fn split_sql_statements_with_options(sql: &str, options: SqlParsingOptions) -> Vec<String> {
let mut splitter = SqlStatementSplitter::with_options(options);
let mut statements = splitter.push_chunk(sql);
statements.extend(splitter.finish());
statements
}
#[derive(Debug, Clone, PartialEq, Eq)]
pub struct SqlStatementRange {
pub text: String,
pub start: usize,
pub end: usize,
}
pub fn find_statement_at_cursor(sql: &str, cursor_pos: usize) -> String {
find_statement_at_cursor_with_options(sql, cursor_pos, SqlParsingOptions::default())
}
pub fn find_statement_at_cursor_for_database(sql: &str, cursor_pos: usize, db_type: DatabaseType) -> String {
if db_type == DatabaseType::SqlServer {
return find_sqlserver_statement_at_cursor(sql, cursor_pos);
}
let mut options = SqlParsingOptions::for_database_type(db_type);
if options.profile.supports_psql_control_commands {
let (sql, _) = preprocess_psql_control_commands(sql, options.profile);
options.profile.supports_psql_control_commands = false;
return find_statement_at_cursor_with_options(&sql, cursor_pos, options);
}
find_statement_at_cursor_with_options(sql, cursor_pos, options)
}
pub fn find_statement_at_cursor_with_options(sql: &str, cursor_pos: usize, options: SqlParsingOptions) -> String {
let statements = split_sql_statement_ranges_with_options(sql, options);
let cursor = utf16_offset_to_byte_index(sql, cursor_pos);
for (idx, statement) in statements.iter().enumerate() {
if cursor > statement.start && cursor < statement.end {
return statement_text_at_cursor(sql, statement, cursor, options);
}
if cursor == statement.start {
if cursor_has_sql_after_cursor_on_line(sql, cursor) {
return statement_text_at_cursor(sql, statement, cursor, options);
}
if let Some(prev) = idx.checked_sub(1).and_then(|prev_idx| statements.get(prev_idx)) {
return statement_text_at_cursor(sql, prev, cursor, options);
}
return statement_text_at_cursor(sql, statement, cursor, options);
}
if cursor < statement.start {
if let Some(prev) = idx.checked_sub(1).and_then(|prev_idx| statements.get(prev_idx)) {
return statement_text_at_cursor(sql, prev, cursor, options);
}
return statement_text_at_cursor(sql, statement, cursor, options);
}
}
statements
.last()
.map(|statement| statement_text_at_cursor(sql, statement, cursor, options))
.unwrap_or_else(|| sql.trim().to_string())
}
fn cursor_has_sql_after_cursor_on_line(sql: &str, cursor: usize) -> bool {
let line_end = sql[cursor..].find('\n').map_or(sql.len(), |offset| cursor + offset);
sql[cursor..line_end].chars().any(|ch| !ch.is_whitespace())
}
fn statement_text_at_cursor(
sql: &str,
statement: &SqlStatementRange,
cursor: usize,
options: SqlParsingOptions,
) -> String {
let soft_ranges = split_statement_range_at_blank_lines(sql, statement, options);
find_statement_text_in_ranges(sql, &soft_ranges, cursor).unwrap_or_else(|| statement.text.clone())
}
fn find_statement_text_in_ranges(sql: &str, ranges: &[SqlStatementRange], cursor: usize) -> Option<String> {
for (idx, range) in ranges.iter().enumerate() {
if cursor > range.start && cursor < range.end {
return Some(range.text.clone());
}
if cursor == range.start {
if cursor_has_sql_after_cursor_on_line(sql, cursor) {
return Some(range.text.clone());
}
if let Some(prev) = idx.checked_sub(1).and_then(|prev_idx| ranges.get(prev_idx)) {
return Some(prev.text.clone());
}
return Some(range.text.clone());
}
if cursor < range.start {
if let Some(prev) = idx.checked_sub(1).and_then(|prev_idx| ranges.get(prev_idx)) {
return Some(prev.text.clone());
}
return Some(range.text.clone());
}
}
ranges.last().map(|range| range.text.clone())
}
fn split_statement_range_at_blank_lines(
sql: &str,
statement: &SqlStatementRange,
options: SqlParsingOptions,
) -> Vec<SqlStatementRange> {
if keeps_oracle_style_block_together(options, &statement.text) {
return vec![statement.clone()];
}
if options.profile.supports_hana_script_blocks && starts_with_hana_script_block(&statement.text) {
return vec![statement.clone()];
}
if options.profile.supports_firebird_routine_blocks && starts_with_firebird_routine_block(&statement.text) {
return vec![statement.clone()];
}
let mut ranges = Vec::new();
let mut scanner = SqlScanner::with_profile(options.profile);
let mut current_start = statement.start;
let mut line_start = statement.start;
let mut line_has_non_whitespace = false;
let mut blank_line_run = 0usize;
for (relative_idx, ch) in sql[statement.start..statement.end].char_indices() {
let idx = statement.start + relative_idx;
if ch == '\n' {
if !line_has_non_whitespace && !scanner.is_masked() {
blank_line_run += 1;
} else {
blank_line_run = 0;
}
scanner.step(sql, idx, ch);
line_start = idx + ch.len_utf8();
line_has_non_whitespace = false;
continue;
}
if !line_has_non_whitespace && !ch.is_whitespace() {
if blank_line_run >= 2
&& !scanner.is_masked()
&& has_executable_sql_with_options(&sql[current_start..line_start], options)
&& starts_with_soft_statement_keyword(&sql[line_start..statement.end], options)
{
push_statement_range(&mut ranges, sql, current_start, line_start, options);
current_start = line_start;
}
blank_line_run = 0;
line_has_non_whitespace = true;
}
scanner.step(sql, idx, ch);
}
push_statement_range(&mut ranges, sql, current_start, statement.end, options);
if ranges.is_empty() {
vec![statement.clone()]
} else {
ranges
}
}
fn starts_with_soft_statement_keyword(sql: &str, options: SqlParsingOptions) -> bool {
if options.profile.supports_hana_script_blocks
&& starts_with_executable_sql_keyword_with_options(sql, &["DO"], options)
{
return true;
}
if options.profile.supports_firebird_routine_blocks
&& starts_with_executable_sql_keyword_with_options(sql, &["RECREATE"], options)
{
return true;
}
starts_with_executable_sql_keyword_with_options(
sql,
&[
"CREATE", "ALTER", "DROP", "INSERT", "UPDATE", "DELETE", "MERGE", "REPLACE", "TRUNCATE", "GRANT", "REVOKE",
"COMMENT", "EXPLAIN", "SHOW", "DESCRIBE", "USE", "SET", "CALL", "EXEC", "EXECUTE", "BEGIN", "COMMIT",
"ROLLBACK", "DECLARE", "ANALYZE", "VACUUM", "PRAGMA", "REFRESH", "COPY",
],
options,
)
}
#[allow(dead_code)]
fn split_sql_statement_ranges(sql: &str) -> Vec<SqlStatementRange> {
split_sql_statement_ranges_with_options(sql, SqlParsingOptions::default())
}
/// Statement ranges that keep their byte offsets into `sql`, so callers can rewrite
/// individual statements in place (see the Oracle administrative DDL tolerance in
/// `sql_analysis`).
pub(crate) fn statement_ranges_for_database(sql: &str, db_type: DatabaseType) -> Vec<SqlStatementRange> {
split_sql_statement_ranges_with_options(sql, SqlParsingOptions::for_database_type(db_type))
}
fn split_sql_statement_ranges_with_options(sql: &str, options: SqlParsingOptions) -> Vec<SqlStatementRange> {
let mut ranges = Vec::new();
let mut start = 0;
let mut i = 0;
let mut in_single_quote = false;
let mut in_double_quote = false;
let mut in_backtick = false;
let mut in_line_comment = false;
let mut block_comment_depth = 0usize;
let mut dollar_quote_tag: Option<String> = None;
let mut custom_delimiter: Option<String> = None;
let mut single_quote_escape_string = false;
let mut postgres_dollar_quoted_routine = false;
let mut oracle_block = OraclePlSqlBlockTracker::default();
let mut oracle_block_start = 0;
while i < sql.len() {
if let Some(tag) = &dollar_quote_tag {
if sql[i..].starts_with(tag) {
i += tag.len();
dollar_quote_tag = None;
continue;
}
i += next_char_len(sql, i);
continue;
}
let ch = next_char(sql, i);
let next = next_char_at(sql, i + ch.len_utf8());
if in_line_comment {
i += ch.len_utf8();
if ch == '\n' {
in_line_comment = false;
}
continue;
}
if block_comment_depth > 0 {
if ch == '*' && next == Some('/') {
i += 2;
block_comment_depth -= 1;
} else if options.profile.supports_nested_block_comments && ch == '/' && next == Some('*') {
i += 2;
block_comment_depth += 1;
} else {
i += ch.len_utf8();
}
continue;
}
if !in_single_quote && !in_double_quote && !in_backtick {
if options.profile.supports_oracle_q_quotes && ch == '\'' && ends_with_oracle_q_quote_prefix(&sql[..i]) {
if let Some(end) = oracle_q_quote_end(sql, i) {
i = end;
continue;
}
}
if ch == '-'
&& next == Some('-')
&& dash_dash_starts_line_comment(options.profile, next_char_at(sql, i + 2))
{
in_line_comment = true;
i += 2;
continue;
}
if options.profile.supports_hash_line_comments && ch == '#' {
in_line_comment = true;
i += ch.len_utf8();
continue;
}
if ch == '/' && next == Some('*') {
block_comment_depth = 1;
i += 2;
continue;
}
if let Some(tag) =
options.profile.supports_dollar_quoted_strings.then(|| dollar_quote_tag_at_str(sql, i)).flatten()
{
if custom_delimiter.is_none() && !is_on_delimiter_line(sql, start, i, options) {
if options.profile.supports_postgres_dollar_quoted_routines
&& starts_with_postgres_dollar_quoted_routine_prefix(&sql[start..i])
{
postgres_dollar_quoted_routine = true;
}
i += tag.len();
dollar_quote_tag = Some(tag);
continue;
}
}
if ch == '\n' {
let line_start = sql[..i].rfind('\n').map_or(0, |pos| pos + 1);
let line = sql[line_start..i].trim();
if options.profile.supports_slash_line_block_delimiter && line == "/" {
push_statement_range(&mut ranges, sql, start, line_start, options);
start = i + ch.len_utf8();
postgres_dollar_quoted_routine = false;
i = start;
continue;
}
if let Some(new_delimiter) = options
.profile
.supports_custom_delimiter_commands
.then(|| {
parse_delimiter_command(line)
.filter(|_| !has_executable_sql_with_options(&sql[start..line_start], options))
})
.flatten()
{
let before = sql[start..line_start].trim();
if has_executable_sql_with_options(before, options) {
push_statement_range(&mut ranges, sql, start, line_start, options);
}
custom_delimiter = if new_delimiter == ";" { None } else { Some(new_delimiter.to_string()) };
start = i + ch.len_utf8();
postgres_dollar_quoted_routine = false;
i = start;
continue;
}
}
}
let single_quote_backslash_escapes = if in_single_quote {
single_quote_escape_string
} else {
options.profile.supports_backslash_escaped_quotes
};
match ch {
'\'' if !(in_double_quote
|| in_backtick
|| single_quote_backslash_escapes && has_odd_trailing_backslashes(&sql[start..i])) =>
{
in_single_quote = !in_single_quote;
single_quote_escape_string = in_single_quote
&& (options.profile.supports_backslash_escaped_quotes
|| (options.profile.supports_postgres_escape_strings
&& ends_with_escape_string_prefix(&sql[start..i])));
i += ch.len_utf8();
}
'"' if !(in_single_quote
|| in_backtick
|| options.profile.supports_backslash_escaped_quotes
&& has_odd_trailing_backslashes(&sql[start..i])) =>
{
in_double_quote = !in_double_quote;
i += ch.len_utf8();
}
'`' if options.profile.supports_backtick_identifiers && !in_single_quote && !in_double_quote => {
in_backtick = !in_backtick;
i += ch.len_utf8();
}
';' if !(in_single_quote
|| in_double_quote
|| in_backtick
|| custom_delimiter.is_some()
|| (options.profile.supports_custom_delimiter_commands
&& is_on_delimiter_line(sql, start, i, options))) =>
{
let is_mysql_routine =
options.profile.supports_mysql_routine_blocks && starts_with_mysql_routine_block(&sql[start..i]);
if is_mysql_routine {
if !mysql_routine_block_is_complete(&sql[start..i + ch.len_utf8()]) {
i += ch.len_utf8();
continue;
}
push_statement_range(&mut ranges, sql, start, i, options);
} else {
if oracle_block_start != start {
oracle_block.reset();
oracle_block_start = start;
}
let is_oracle_plsql =
!postgres_dollar_quoted_routine && oracle_block.keeps_block_together(options, &sql[start..i]);
if is_oracle_plsql {
if !oracle_block.is_complete(&sql[start..i + ch.len_utf8()]) {
i += ch.len_utf8();
continue;
}
push_statement_range(&mut ranges, sql, start, i + ch.len_utf8(), options);
} else if options.profile.supports_hana_script_blocks
&& starts_with_hana_script_block(&sql[start..i])
{
if !hana_script_block_is_complete(&sql[start..i + ch.len_utf8()]) {
i += ch.len_utf8();
continue;
}
push_statement_range(&mut ranges, sql, start, i + ch.len_utf8(), options);
} else if options.profile.supports_firebird_routine_blocks
&& starts_with_firebird_routine_block(&sql[start..i])
{
if !firebird_routine_block_is_complete(&sql[start..i + ch.len_utf8()]) {
i += ch.len_utf8();
continue;
}
push_statement_range(&mut ranges, sql, start, i + ch.len_utf8(), options);
} else if keeps_sqlserver_control_flow_batch(options, &sql[start..i]) {
// T-SQL does not end a control-flow batch at a semicolon
// (#9336, #13051): the inner `;` stays in the batch and
// the statement ends at the block's final `END [TRY]`.
if !sqlserver_control_flow_batch_is_complete(&sql[start..i + ch.len_utf8()]) {
i += ch.len_utf8();
continue;
}
// A final `END` without its own `;` shares the fragment
// with the next statement; the boundary sits right after
// that `END` instead (TS parity with
// `sqlStatementRanges.ts`). An `END` directly followed
// by this `;` keeps the historical range shape.
let body = &sql[start..i];
let mut boundaries = Vec::new();
let mut cursor = 0usize;
while let Some(relative) = sqlserver_control_flow_batch_boundary(&body[cursor..]) {
if cursor + relative >= body.len() {
break;
}
cursor += relative;
boundaries.push(cursor);
}
if boundaries.is_empty() {
push_statement_range(&mut ranges, sql, start, i + ch.len_utf8(), options);
} else {
let mut piece_from = 0usize;
for boundary_end in &boundaries {
push_statement_range(
&mut ranges,
sql,
start + piece_from,
start + boundary_end,
options,
);
piece_from = *boundary_end;
}
let trailing = &body[piece_from..];
if !trailing.trim().is_empty() {
if keeps_sqlserver_control_flow_batch(options, trailing) {
if sqlserver_control_flow_batch_is_complete(trailing) {
push_statement_range(
&mut ranges,
sql,
start + piece_from,
i + ch.len_utf8(),
options,
);
} else {
// The remainder opens a new batch that is
// still mid-typing: this `;` stays inside it.
start += piece_from;
i += ch.len_utf8();
continue;
}
} else {
push_statement_range(&mut ranges, sql, start + piece_from, i, options);
}
}
}
} else {
push_statement_range(&mut ranges, sql, start, i, options);
}
}
i += ch.len_utf8();
start = i;
postgres_dollar_quoted_routine = false;
}
_ => {
i += ch.len_utf8();
if !in_single_quote && !in_double_quote && !in_backtick {
if let Some(delimiter) = &custom_delimiter {
if sql[start..i].ends_with(delimiter) && !is_on_delimiter_line(sql, start, i, options) {
let end = i - delimiter.len();
push_statement_range(&mut ranges, sql, start, end, options);
start = i;
postgres_dollar_quoted_routine = false;
}
}
}
}
}
}
let trimmed = sql[start..].trim();
let last_line = trimmed.rsplit('\n').next().unwrap_or(trimmed).trim();
let before_last_line = trimmed.rsplit_once('\n').map(|x| x.0).unwrap_or("").trim();
if options.profile.supports_custom_delimiter_commands
&& parse_delimiter_command(last_line)
.is_some_and(|_| !has_executable_sql_with_options(before_last_line, options))
{
if let Some(line_start) = sql[start..].rfind('\n').map(|pos| start + pos + 1) {
push_statement_range(&mut ranges, sql, start, line_start, options);
}
} else if options.profile.supports_slash_line_block_delimiter && last_line == "/" {
if let Some(line_start) = sql[start..].rfind('\n').map(|pos| start + pos + 1) {
push_statement_range(&mut ranges, sql, start, line_start, options);
}
} else {
push_statement_range(&mut ranges, sql, start, sql.len(), options);
}
ranges
}
fn push_statement_range(
ranges: &mut Vec<SqlStatementRange>,
sql: &str,
start: usize,
end: usize,
options: SqlParsingOptions,
) {
let Some((relative_start, relative_end)) = executable_sql_bounds(&sql[start..end], options) else {
return;
};
let statement_start = start + relative_start;
let statement_end = start + relative_end;
let text = sql[statement_start..statement_end].to_string();
if !text.is_empty() {
ranges.push(SqlStatementRange { text, start: statement_start, end: statement_end });
}
}
fn utf16_offset_to_byte_index(sql: &str, offset: usize) -> usize {
let mut utf16_seen = 0;
for (byte_index, ch) in sql.char_indices() {
if utf16_seen >= offset {
return byte_index;
}
utf16_seen += ch.len_utf16();
if utf16_seen > offset {
return byte_index + ch.len_utf8();
}
}
sql.len()
}
fn next_char(sql: &str, index: usize) -> char {
sql[index..].chars().next().unwrap_or('\0')
}
fn next_char_at(sql: &str, index: usize) -> Option<char> {
if index >= sql.len() {
None
} else {
sql[index..].chars().next()
}
}
fn next_char_len(sql: &str, index: usize) -> usize {
next_char(sql, index).len_utf8()
}
fn is_on_delimiter_line(sql: &str, range_start: usize, index: usize, options: SqlParsingOptions) -> bool {
let line_start = sql[range_start..index].rfind('\n').map_or(range_start, |pos| range_start + pos + 1);
sql[line_start..index].trim_start().as_bytes().get(..9).is_some_and(|prefix| {
prefix.eq_ignore_ascii_case(b"delimiter")
&& !has_executable_sql_with_options(sql[range_start..line_start].trim(), options)
})
}
fn dollar_quote_tag_at_str(sql: &str, index: usize) -> Option<String> {
let rest = &sql[index..];
if !rest.starts_with('$') {
return None;
}
let end = rest[1..].find('$')? + 1;
let tag = &rest[..=end];
if tag.len() == 2 {
return Some(tag.to_string());
}
let name = &tag[1..tag.len() - 1];
if !name.chars().all(|ch| ch == '_' || ch.is_ascii_alphanumeric()) {
return None;
}
Some(tag.to_string())
}
pub fn split_sql_batches(sql: &str) -> Vec<String> {
let ranges = split_sql_batch_ranges(sql, SqlDialectProfile::sql_server());
if ranges.is_empty() {
let trimmed = sql.trim();
return if trimmed.is_empty() { Vec::new() } else { vec![trimmed.to_string()] };
}
ranges.into_iter().map(|range| range.text).collect()
}
fn split_sql_batch_ranges(sql: &str, profile: SqlDialectProfile) -> Vec<SqlStatementRange> {
let mut batches = Vec::new();
let mut current_start = 0;
let lines: Vec<&str> = sql.split('\n').collect();
let mut offset = 0;
let mut scanner = SqlScanner::with_profile(profile);
for line in &lines {
let line_start = offset;
let line_end = offset + line.len();
offset = line_end + 1; // +1 for the '\n'
let trimmed = line.trim();
let is_batch_separator = profile.supports_go_batch_separator
&& !scanner.is_masked()
&& (trimmed.eq_ignore_ascii_case("go")
|| trimmed.to_ascii_lowercase().starts_with("go ") && trimmed[2..].trim().is_empty());
if is_batch_separator {
push_batch_range(&mut batches, sql, current_start, line_start);
current_start = line_end.min(sql.len());
if current_start < sql.len() && sql.as_bytes()[current_start] == b'\n' {
current_start += 1;
}
} else {
for (relative_idx, ch) in line.char_indices() {
scanner.step(sql, line_start + relative_idx, ch);
}
}
if line_end < sql.len() {
scanner.step(sql, line_end, '\n');
}
}
push_batch_range(&mut batches, sql, current_start, sql.len());
batches
}
fn push_batch_range(ranges: &mut Vec<SqlStatementRange>, sql: &str, start: usize, end: usize) {
let Some((relative_start, relative_end)) = executable_sql_bounds(&sql[start..end], SqlParsingOptions::default())
else {
return;
};
let statement_start = start + relative_start;
let statement_end = start + relative_end;
let text = sql[statement_start..statement_end].to_string();
if !text.is_empty() {
ranges.push(SqlStatementRange { text, start: statement_start, end: statement_end });
}
}
fn find_sqlserver_statement_at_cursor(sql: &str, cursor_pos: usize) -> String {
let profile = SqlDialectProfile::sql_server();
let cursor = utf16_offset_to_byte_index(sql, cursor_pos);
let batches = split_sql_batch_ranges(sql, profile);
for (idx, batch) in batches.iter().enumerate() {
if cursor >= batch.start && cursor <= batch.end {
// Module DDL is checked before the block scan: a procedure body is
// itself a `BEGIN ... END` block, and the whole `CREATE`/`ALTER`
// batch is the execution unit there.
if profile.keeps_sqlserver_module_batch_at_cursor && starts_with_sqlserver_module_ddl(&batch.text) {
return batch.text.clone();
}
if let Some(statement) =
sqlserver_control_flow_statement_at(&batch.text, Some(cursor.saturating_sub(batch.start)))
{
return statement;
}
let relative_cursor = sql[..cursor].encode_utf16().count() - sql[..batch.start].encode_utf16().count();
return find_statement_at_cursor_with_options(&batch.text, relative_cursor, SqlParsingOptions::default());
}
if cursor < batch.start {
if let Some(prev) = idx.checked_sub(1).and_then(|prev_idx| batches.get(prev_idx)) {
if profile.keeps_sqlserver_module_batch_at_cursor && starts_with_sqlserver_module_ddl(&prev.text) {
return prev.text.clone();
}
if let Some(statement) = sqlserver_control_flow_statement_at(&prev.text, None) {
return statement;
}
let relative_cursor = prev.text.encode_utf16().count();
return find_statement_at_cursor_with_options(
&prev.text,
relative_cursor,
SqlParsingOptions::default(),
);
}
return batch.text.clone();
}
}
batches.last().map(|batch| batch.text.clone()).unwrap_or_else(|| sql.trim().to_string())
}
/// A top-level token of a SQL Server batch: a word or a `;` terminator outside
/// string literals, quoted identifiers, comments and parentheses. `CASE ... END`
/// inside a condition or select list therefore never looks like a block.
struct SqlServerTopLevelToken<'a> {
text: &'a str,
start: usize,
end: usize,
}
impl SqlServerTopLevelToken<'_> {
fn is_word(&self, keyword: &str) -> bool {
self.text.eq_ignore_ascii_case(keyword)
}
}
fn sqlserver_top_level_tokens(sql: &str) -> Vec<SqlServerTopLevelToken<'_>> {
let bytes = sql.as_bytes();
let mut tokens = Vec::new();
let mut paren_depth = 0usize;
let mut index = 0usize;
while index < bytes.len() {
match bytes[index] {
quote @ (b'\'' | b'"') => {
index += 1;
while index < bytes.len() {
if bytes[index] == quote {
if bytes.get(index + 1) == Some("e) {
index += 2;
continue;
}
index += 1;
break;
}
index += 1;
}
}
b'[' => {
index += 1;
while index < bytes.len() {
if bytes[index] == b']' {
if bytes.get(index + 1) == Some(&b']') {
index += 2;
continue;
}
index += 1;
break;
}
index += 1;
}
}
b'-' if bytes.get(index + 1) == Some(&b'-') => {
index += 2;
while index < bytes.len() && bytes[index] != b'\n' {
index += 1;
}
}
b'/' if bytes.get(index + 1) == Some(&b'*') => {
index += 2;
while index < bytes.len() && !(bytes[index] == b'*' && bytes.get(index + 1) == Some(&b'/')) {
index += 1;
}
index = (index + 2).min(bytes.len());
}
b'(' => {
paren_depth += 1;
index += 1;
}
b')' => {
paren_depth = paren_depth.saturating_sub(1);
index += 1;
}
b';' if paren_depth == 0 => {
tokens.push(SqlServerTopLevelToken { text: ";", start: index, end: index + 1 });
index += 1;
}
byte if paren_depth == 0 && (byte == b'_' || byte.is_ascii_alphabetic()) => {
let start = index;
index += 1;
while index < bytes.len() {
let next = bytes[index];
if next == b'_' || next.is_ascii_alphanumeric() || next == b'$' {
index += 1;
} else {
break;
}
}
tokens.push(SqlServerTopLevelToken { text: &sql[start..index], start, end: index });
}
_ => index += 1,
}
}
tokens
}
/// `tokens[index]` opens a real block (`BEGIN ... END`, `BEGIN TRY`,
/// `BEGIN CATCH`) rather than a transaction or conversation, which has no
/// matching `END`.
fn sqlserver_begins_block(tokens: &[SqlServerTopLevelToken<'_>], index: usize) -> bool {
let Some(next) = tokens.get(index + 1) else {
return false;
};
next.text != ";"
&& !["TRANSACTION", "TRAN", "DISTRIBUTED", "DIALOG", "CONVERSATION"].iter().any(|keyword| next.is_word(keyword))
}
#[derive(Clone, Copy, PartialEq, Eq)]
enum SqlServerBlockKind {
Begin,
Case,
}
struct SqlServerBlockFrame {
kind: SqlServerBlockKind,
start: usize,
/// An `IF`/`WHILE`/`ELSE` appeared since the last top-level `;`, so this
/// block is a branch body and its control statement is the execution unit.
control_statement: bool,
}
/// Byte span of a top-level SQL Server `BEGIN ... END` block.
#[derive(Clone, Copy)]
struct SqlServerControlFlowBlock {
start: usize,
end: usize,
control_statement: bool,
}
/// Top-level `BEGIN ... END` blocks of a batch (`BEGIN TRY ... END TRY BEGIN
/// CATCH ... END CATCH` counts as one). `CASE` frames keep their own `END` from
/// closing the surrounding block, and an unclosed block runs to the end of the
/// batch so a block being typed stays whole.
fn sqlserver_top_level_blocks(sql: &str) -> Vec<SqlServerControlFlowBlock> {
let tokens = sqlserver_top_level_tokens(sql);
let mut blocks = Vec::new();
let mut stack: Vec<SqlServerBlockFrame> = Vec::new();
let mut pending_try: Option<SqlServerBlockFrame> = None;
let mut in_control_statement = false;
for (index, token) in tokens.iter().enumerate() {
if token.text == ";" {
in_control_statement = false;
continue;
}
if token.is_word("IF") || token.is_word("WHILE") || token.is_word("ELSE") {
in_control_statement = true;
continue;
}
if token.is_word("BEGIN") {
if !sqlserver_begins_block(&tokens, index) {
continue;
}
let frame = match pending_try.take() {
Some(frame) if stack.is_empty() => frame,
_ => SqlServerBlockFrame {
kind: SqlServerBlockKind::Begin,
start: token.start,
control_statement: in_control_statement,
},
};
stack.push(frame);
continue;
}
if token.is_word("CASE") {
stack.push(SqlServerBlockFrame {
kind: SqlServerBlockKind::Case,
start: token.start,
control_statement: in_control_statement,
});
continue;
}
if !token.is_word("END") {
continue;
}
let next = tokens.get(index + 1);
// `END CONVERSATION`/`END DIALOG` close a conversation, not a block.
if next.is_some_and(|next| next.is_word("CONVERSATION") || next.is_word("DIALOG")) {
continue;
}
let Some(frame) = stack.pop() else {
continue;
};
if frame.kind != SqlServerBlockKind::Begin || !stack.is_empty() {
continue;
}
// `END TRY`/`END CATCH` are two-token closers: the span includes the
// keyword so the boundary sits after the pair.
let end = match next {
Some(next) if next.is_word("TRY") || next.is_word("CATCH") => next.end,
_ => token.end,
};
if next.is_some_and(|next| next.is_word("TRY"))
&& tokens.get(index + 2).is_some_and(|token| token.is_word("BEGIN"))
&& tokens.get(index + 3).is_some_and(|token| token.is_word("CATCH"))
{
// `END TRY` immediately followed by `BEGIN CATCH`: both halves are
// one batch, so the CATCH block continues the TRY block's span.
pending_try = Some(frame);
continue;
}
blocks.push(SqlServerControlFlowBlock { start: frame.start, end, control_statement: frame.control_statement });
}
if let Some(frame) = pending_try {
// `END TRY` without its `BEGIN CATCH` yet: keep the half written batch
// whole while the user is still typing it.
blocks.push(SqlServerControlFlowBlock {
start: frame.start,
end: sql.len(),
control_statement: frame.control_statement,
});
} else if let Some(frame) = stack.first() {
if frame.kind == SqlServerBlockKind::Begin {
blocks.push(SqlServerControlFlowBlock {
start: frame.start,
end: sql.len(),
control_statement: frame.control_statement,
});
}
}
blocks
}
/// Text of the SQL Server control-flow statement that contains `cursor` (a byte
/// offset into `sql`), or `None` when the batch is ordinary `;`-separated
/// statements. `None` for the cursor means it sits after this batch, so any
/// top-level block keeps the whole batch.
///
/// `IF`/`WHILE` batches and top-level `BEGIN ... END` blocks contain no
/// independent statements: a `;` inside them does not end an execution range
/// (#9336), and a batch that only *contains* the block, after a leading
/// `SET XACT_ABORT ON;` for instance, still has to run it as one unit (#13051).
fn sqlserver_control_flow_statement_at(sql: &str, cursor: Option<usize>) -> Option<String> {
if starts_with_sqlserver_control_flow_batch(sql) {
return Some(sql.to_string());
}
let blocks = sqlserver_top_level_blocks(sql);
let block = match cursor {
Some(cursor) => blocks.iter().find(|block| cursor >= block.start && cursor <= block.end),
None => blocks.last(),
}?;
// A branch body alone is invalid T-SQL, so its whole control statement is
// the unit; the whole batch keeps the leading statements too.
if block.control_statement || cursor.is_none() {
return Some(sql.to_string());
}
Some(sql[block.start..block.end].to_string())
}
/// True when the batch starts with a whole T-SQL control-flow statement: an
/// `IF` batch, or a `BEGIN` block. `BEGIN` keywords that start a transaction or
/// conversation have no matching `END` and stay split per statement.
fn starts_with_sqlserver_control_flow_batch(sql: &str) -> bool {
let tokens = sqlserver_top_level_tokens(sql);
let Some(first) = tokens.first() else {
return false;
};
if first.is_word("IF") {
return tokens.iter().any(|token| token.is_word("ELSE"))
&& tokens.iter().any(|token| token.is_word("BEGIN"))
&& tokens.iter().any(|token| token.is_word("END"));
}
first.is_word("BEGIN") && sqlserver_begins_block(&tokens, 0) && tokens.iter().any(|token| token.is_word("END"))
}
pub fn starts_with_sqlserver_module_ddl(sql: &str) -> bool {
let tokens = first_sql_tokens(sql, 4);
if tokens.len() >= 4
&& tokens[0].eq_ignore_ascii_case("CREATE")
&& tokens[1].eq_ignore_ascii_case("OR")
&& tokens[2].eq_ignore_ascii_case("ALTER")
{
return is_sqlserver_module_keyword(&tokens[3]);
}
tokens.len() >= 2
&& (tokens[0].eq_ignore_ascii_case("CREATE") || tokens[0].eq_ignore_ascii_case("ALTER"))
&& is_sqlserver_module_keyword(&tokens[1])
}
fn is_sqlserver_module_keyword(token: &str) -> bool {
["FUNCTION", "PROC", "PROCEDURE", "TRIGGER", "VIEW"].iter().any(|keyword| token.eq_ignore_ascii_case(keyword))
}
fn starts_with_mysql_routine_block(sql: &str) -> bool {
is_mysql_routine_ddl_start(sql) && mysql_routine_tokens(sql).iter().any(|token| token.eq_ignore_ascii_case("BEGIN"))
}
fn is_mysql_routine_ddl_start(sql: &str) -> bool {
let executable = leading_executable_sql_with_options(sql, SqlParsingOptions::mysql_compatible());
let tokens = first_sql_tokens(executable, 16);
if tokens.first().is_none_or(|token| !token.eq_ignore_ascii_case("CREATE")) {
return false;
}
for token in tokens.iter().skip(1) {
if ["PROCEDURE", "FUNCTION", "TRIGGER", "EVENT"].iter().any(|keyword| token.eq_ignore_ascii_case(keyword)) {
return true;
}
if [
"DATABASE",
"INDEX",
"LOGFILE",
"ROLE",
"SCHEMA",
"SERVER",
"SPATIAL",
"TABLE",
"TEMPORARY",
"UNIQUE",
"USER",
"VIEW",
]
.iter()
.any(|keyword| token.eq_ignore_ascii_case(keyword))
{
return false;
}
}
false
}
#[derive(Clone, Copy, PartialEq, Eq)]
enum MysqlRoutineBlockType {
Begin,
Case,
}
fn mysql_routine_block_is_complete(sql: &str) -> bool {
if !starts_with_mysql_routine_block(sql) {
return false;
}
let tokens = mysql_routine_tokens(sql);
let mut block_stack = Vec::new();
let mut saw_begin = false;
for (index, token) in tokens.iter().enumerate() {
if token == ";" {
continue;
}
if token.eq_ignore_ascii_case("BEGIN") {
if previous_mysql_routine_word(&tokens, index).is_some_and(|previous| previous.eq_ignore_ascii_case("END"))
{
continue;
}
saw_begin = true;
block_stack.push(MysqlRoutineBlockType::Begin);
continue;
}
if token.eq_ignore_ascii_case("CASE") {
if previous_mysql_routine_word(&tokens, index).is_some_and(|previous| previous.eq_ignore_ascii_case("END"))
{
continue;
}
block_stack.push(MysqlRoutineBlockType::Case);
continue;
}
if token.eq_ignore_ascii_case("END") && saw_begin {
let suffix = next_mysql_routine_word(&tokens, index);
if suffix.is_some_and(|next| next.eq_ignore_ascii_case("CASE")) {
if block_stack.last() == Some(&MysqlRoutineBlockType::Case) {
block_stack.pop();
}
continue;
}
if suffix.is_some_and(is_mysql_control_block_suffix) {
continue;
}
block_stack.pop();
}
}
saw_begin && block_stack.is_empty() && tokens.last().is_some_and(|token| token == ";")
}
fn is_mysql_control_block_suffix(token: &str) -> bool {
["IF", "LOOP", "CASE", "REPEAT", "WHILE"].iter().any(|keyword| token.eq_ignore_ascii_case(keyword))
}
fn previous_mysql_routine_word(tokens: &[String], index: usize) -> Option<&str> {
tokens[..index].iter().rev().find(|token| token.as_str() != ";").map(String::as_str)
}
fn next_mysql_routine_word(tokens: &[String], index: usize) -> Option<&str> {
tokens.get(index + 1..)?.iter().find(|token| token.as_str() != ";").map(String::as_str)
}
fn mysql_routine_tokens(sql: &str) -> Vec<String> {
let chars = sql.chars().collect::<Vec<_>>();
let mut tokens = Vec::new();
let mut in_line_comment = false;
let mut in_block_comment = false;
let mut in_single_quote = false;
let mut in_double_quote = false;
let mut in_backtick = false;
let mut i = 0;
while i < chars.len() {
let ch = chars[i];
let next = chars.get(i + 1).copied();
if in_line_comment {
if ch == '\n' {
in_line_comment = false;
}
i += 1;
continue;
}
if in_block_comment {
// MySQL block comments do not nest, so a single flag is enough here.
if ch == '*' && next == Some('/') {
in_block_comment = false;
i += 2;
} else {
i += 1;
}
continue;
}
if in_single_quote {
if ch == '\\' && next.is_some() {
i += 2;
continue;
}
if ch == '\'' {
if next == Some('\'') {
i += 2;
continue;
}
in_single_quote = false;
}
i += 1;
continue;
}
if in_double_quote {
if ch == '\\' && next.is_some() {
i += 2;
continue;
}
if ch == '"' {
if next == Some('"') {
i += 2;
continue;
}
in_double_quote = false;
}
i += 1;
continue;
}
if in_backtick {
if ch == '`' {
if next == Some('`') {
i += 2;
continue;
}
in_backtick = false;
}
i += 1;
continue;
}
if ch == '-' && next == Some('-') {
in_line_comment = true;
i += 2;
continue;
}
if ch == '#' {
in_line_comment = true;
i += 1;
continue;
}
if ch == '/' && next == Some('*') {
in_block_comment = true;
i += 2;
continue;
}
if ch == '\'' {
in_single_quote = true;
i += 1;
continue;
}
if ch == '"' {
in_double_quote = true;
i += 1;
continue;
}
if ch == '`' {
in_backtick = true;
i += 1;
continue;
}
if ch == ';' {
tokens.push(";".to_string());
i += 1;
continue;
}
if ch == '_' || ch.is_ascii_alphabetic() {
let start = i;
i += 1;
while i < chars.len() && is_sql_ident_char(chars[i]) {
i += 1;
}
tokens.push(chars[start..i].iter().collect::<String>().to_ascii_uppercase());
continue;
}
i += 1;
}
tokens
}
fn first_sql_tokens(sql: &str, limit: usize) -> Vec<String> {
let mut tokens = Vec::new();
let mut token = String::new();
let mut chars = sql.chars().peekable();
while let Some(ch) = chars.next() {
if ch == '-' && chars.peek() == Some(&'-') {
chars.next();
if !token.is_empty() {
tokens.push(std::mem::take(&mut token));
}
for comment_char in chars.by_ref() {
if comment_char == '\n' {
break;
}
}
} else if ch == '/' && chars.peek() == Some(&'*') {
chars.next();
if !token.is_empty() {
tokens.push(std::mem::take(&mut token));
}
let mut previous = None;
for comment_char in chars.by_ref() {
if previous == Some('*') && comment_char == '/' {
break;
}
previous = Some(comment_char);
}
} else if ch.is_ascii_alphanumeric() || ch == '_' {
token.push(ch);
} else if !token.is_empty() {
tokens.push(std::mem::take(&mut token));
}
if tokens.len() >= limit {
break;
}
}
if tokens.len() < limit && !token.is_empty() {
tokens.push(token);
}
tokens
}
fn parse_delimiter_command(line: &str) -> Option<&str> {
let line = line.trim();
if line.eq_ignore_ascii_case("delimiter;") {
return Some(";");
}
let bytes = line.as_bytes();
let rest = if bytes.len() > 10
&& (bytes[..10].eq_ignore_ascii_case(b"delimiter ") || bytes[..10].eq_ignore_ascii_case(b"delimiter\t"))
{
Some(&line[10..])
} else {
None
};
rest.map(|r| r.trim()).filter(|r| !r.is_empty())
}
#[derive(Debug, Clone, Copy, PartialEq, Eq)]
enum PsqlControlCommand {
OnErrorStop,
ClientDisplay,
}
fn parse_ignorable_psql_control_command(line: &str) -> Option<PsqlControlCommand> {
let mut parts = line.split_whitespace();
let command = parts.next()?;
if command.eq_ignore_ascii_case("\\echo") {
// Display-only output can be omitted, but another meta-command on the
// same line or psql's backtick command substitution must not be ignored.
let arguments = &line[line.find(command)? + command.len()..];
return (!arguments.contains(['\\', '`'])).then_some(PsqlControlCommand::ClientDisplay);
}
if command.eq_ignore_ascii_case("\\timing") {
let valid =
parts.next().is_none_or(|value| value.eq_ignore_ascii_case("on") || value.eq_ignore_ascii_case("off"));
return (valid && parts.next().is_none()).then_some(PsqlControlCommand::ClientDisplay);
}
if !command.eq_ignore_ascii_case("\\set") {
return None;
}
let variable = parts.next()?;
let value = parts.next()?;
if parts.next().is_some() {
return None;
}
match variable.to_ascii_uppercase().as_str() {
"ON_ERROR_STOP" if matches!(value.to_ascii_lowercase().as_str(), "on" | "true" | "1") => {
Some(PsqlControlCommand::OnErrorStop)
}
"VERBOSITY" if matches!(value.to_ascii_lowercase().as_str(), "default" | "verbose" | "terse" | "sqlstate") => {
Some(PsqlControlCommand::ClientDisplay)
}
_ => None,
}
}
fn preprocess_psql_control_commands(sql: &str, profile: SqlDialectProfile) -> (String, bool) {
let mut output = String::with_capacity(sql.len());
let mut scanner = SqlScanner::with_profile(profile);
let mut offset = 0usize;
let mut stop_on_error = false;
for segment in sql.split_inclusive('\n') {
let line = segment.strip_suffix('\n').unwrap_or(segment);
let command = (!scanner.is_masked()).then(|| parse_ignorable_psql_control_command(line.trim())).flatten();
if let Some(command) = command {
stop_on_error |= command == PsqlControlCommand::OnErrorStop;
// Cursor positions arrive as UTF-16 offsets from the editor. Keep
// those offsets stable even when display text contains emoji.
for ch in line.chars() {
if ch.is_whitespace() {
output.push(ch);
} else {
output.extend(std::iter::repeat_n(' ', ch.len_utf16()));
}
}
if segment.ends_with('\n') {
output.push('\n');
scanner.step(sql, offset + line.len(), '\n');
}
} else {
output.push_str(segment);
for (relative_idx, ch) in segment.char_indices() {
scanner.step(sql, offset + relative_idx, ch);
}
}
offset += segment.len();
}
(output, stop_on_error)
}
pub fn statement_summary(statement: &str) -> String {
const MAX_LEN: usize = 120;
let collapsed = statement.split_whitespace().collect::<Vec<_>>().join(" ");
if collapsed.chars().count() <= MAX_LEN {
return collapsed;
}
collapsed.chars().take(MAX_LEN).collect()
}
pub fn prepare_sql_file_statement(
statement: &str,
db_type: &DatabaseType,
driver_profile: Option<&str>,
) -> SqlFileStatementAction {
let statement = statement.trim();
let postgres_statement =
(*db_type == DatabaseType::Postgres).then(|| strip_postgres_pg_dump_guard_command(statement)).flatten();
let statement = postgres_statement.as_deref().unwrap_or(statement).trim();
if statement.is_empty()
|| (*db_type == DatabaseType::Postgres
&& postgres_statement.is_some()
&& !has_executable_sql_with_options(
statement,
SqlParsingOptions::for_database_type(DatabaseType::Postgres),
))
{
return SqlFileStatementAction::Skip;
}
let is_mysql_compatible_target = is_mysql_compatible_import_target(db_type, driver_profile);
if is_mysql_compatible_target && is_mysql_lock_table_statement(statement) {
return SqlFileStatementAction::Skip;
}
let Some(body) = mysql_executable_comment_body(statement) else {
if is_mysql_compatible_target && is_mysql_session_restore_statement(statement) {
return SqlFileStatementAction::Skip;
}
return SqlFileStatementAction::Execute(statement.to_string());
};
if !is_mysql_compatible_target {
return SqlFileStatementAction::Skip;
}
let body = body.trim();
if body.is_empty() || is_mysql_key_toggle_statement(body) || is_mysql_session_restore_statement(body) {
return SqlFileStatementAction::Skip;
}
SqlFileStatementAction::Execute(body.to_string())
}
fn strip_postgres_pg_dump_guard_command(statement: &str) -> Option<String> {
let executable =
leading_executable_sql_with_options(statement, SqlParsingOptions::for_database_type(DatabaseType::Postgres));
let line_end = executable.find('\n').unwrap_or(executable.len());
if !is_postgres_pg_dump_guard_command(&executable[..line_end]) {
return None;
}
let prefix_len = statement.len() - executable.len();
let suffix_start = prefix_len + line_end + usize::from(line_end < executable.len());
let mut prepared = String::with_capacity(statement.len());
prepared.push_str(&statement[..prefix_len]);
prepared.push_str(&statement[suffix_start..]);
Some(prepared)
}
fn is_postgres_pg_dump_guard_command(line: &str) -> bool {
let mut parts = line.split_whitespace();
let Some(command) = parts.next() else {
return false;
};
if !matches!(command, "\\restrict" | "\\unrestrict") {
return false;
}
let Some(key) = parts.next() else {
return false;
};
parts.next().is_none() && key.bytes().all(|byte| byte.is_ascii_alphanumeric())
}
pub fn optimize_sql_file_import_statements(
statements: &[String],
db_type: Option<DatabaseType>,
driver_profile: Option<&str>,
) -> Vec<SqlFileImportStatement> {
optimize_sql_file_import_statements_with_max_insert_batch_statements(
statements,
db_type,
driver_profile,
SQL_FILE_INSERT_BATCH_MAX_STATEMENTS,
)
}
pub fn optimize_sql_file_import_statements_with_max_insert_batch_statements(
statements: &[String],
db_type: Option<DatabaseType>,
driver_profile: Option<&str>,
max_insert_batch_statements: usize,
) -> Vec<SqlFileImportStatement> {
let mut optimized = Vec::new();
let mut pending_insert: Option<PendingInsertBatch> = None;
let merge_adjacent_inserts =
db_type.as_ref().is_none_or(|db_type| !is_mysql_compatible_import_target(db_type, driver_profile));
let insert_batch_syntax = insert_batch_syntax(db_type.as_ref());
for statement in statements {
let action = db_type
.as_ref()
.map(|db_type| prepare_sql_file_statement(statement, db_type, driver_profile))
.unwrap_or_else(|| SqlFileStatementAction::Execute(statement.trim().to_string()));
match action {
SqlFileStatementAction::Skip => {
flush_pending_insert(&mut optimized, &mut pending_insert);
optimized.push(SqlFileImportStatement {
kind: SqlFileImportStatementKind::Skip,
sql: statement.trim().to_string(),
source_sqls: vec![statement.trim().to_string()],
source_statement_count: 1,
});
}
SqlFileStatementAction::Execute(sql) => {
// Combining separate MySQL INSERT statements changes session
// semantics such as LAST_INSERT_ID(), so execute them with the
// same statement boundaries as the source file.
let mergeable_insert = merge_adjacent_inserts
.then(|| {
let options = db_type.map(SqlParsingOptions::for_database_type).unwrap_or_default();
parse_mergeable_insert(&sql, options)
})
.flatten();
if let Some(insert) = mergeable_insert {
match pending_insert.as_mut() {
Some(batch) if batch.can_accept(&insert, max_insert_batch_statements) => batch.push(insert),
Some(_) => {
flush_pending_insert(&mut optimized, &mut pending_insert);
pending_insert = Some(PendingInsertBatch::new(insert, insert_batch_syntax));
}
None => {
pending_insert = Some(PendingInsertBatch::new(insert, insert_batch_syntax));
}
}
} else {
flush_pending_insert(&mut optimized, &mut pending_insert);
optimized.push(SqlFileImportStatement {
kind: SqlFileImportStatementKind::Execute,
sql: sql.clone(),
source_sqls: vec![sql],
source_statement_count: 1,
});
}
}
}
}
flush_pending_insert(&mut optimized, &mut pending_insert);
optimized
}
fn flush_pending_insert(optimized: &mut Vec<SqlFileImportStatement>, pending_insert: &mut Option<PendingInsertBatch>) {
if let Some(batch) = pending_insert.take() {
optimized.push(SqlFileImportStatement {
kind: SqlFileImportStatementKind::Execute,
sql: batch.to_sql(),
source_sqls: batch.source_sqls,
source_statement_count: batch.source_statement_count,
});
}
}
const SQL_FILE_INSERT_BATCH_MAX_STATEMENTS: usize = 500;
const SQL_FILE_INSERT_BATCH_MAX_BYTES: usize = 4 * 1024 * 1024;
/// How a merged insert batch must be spelled for its target dialect.
///
/// Oracle-family servers reject the MySQL multi-row `VALUES (…), (…)` form, so
/// merging there has to produce `INSERT ALL … SELECT 1 FROM dual` instead.
#[derive(Debug, Clone, Copy, PartialEq, Eq)]
enum InsertBatchSyntax {
MultiRowValues,
OracleInsertAll,
}
fn insert_batch_syntax(db_type: Option<&DatabaseType>) -> InsertBatchSyntax {
match db_type {
Some(DatabaseType::Oracle | DatabaseType::OceanbaseOracle) => InsertBatchSyntax::OracleInsertAll,
_ => InsertBatchSyntax::MultiRowValues,
}
}
#[derive(Debug, Clone)]
struct MergeableInsert {
prefix: String,
prefix_key: String,
values: String,
sql: String,
}
#[derive(Debug, Clone)]
struct PendingInsertBatch {
prefix: String,
prefix_key: String,
values: Vec<String>,
source_sqls: Vec<String>,
source_statement_count: usize,
byte_len: usize,
syntax: InsertBatchSyntax,
}
impl PendingInsertBatch {
fn new(insert: MergeableInsert, syntax: InsertBatchSyntax) -> Self {
let byte_len = insert.prefix.len() + insert.values.len() + 16;
Self {
syntax,
prefix: insert.prefix,
prefix_key: insert.prefix_key,
values: vec![insert.values],
source_sqls: vec![insert.sql],
source_statement_count: 1,
byte_len,
}
}
fn can_accept(&self, insert: &MergeableInsert, max_insert_batch_statements: usize) -> bool {
self.prefix_key == insert.prefix_key
&& self.source_statement_count < max_insert_batch_statements
&& self.byte_len + insert.values.len() + 3 <= SQL_FILE_INSERT_BATCH_MAX_BYTES
}
fn push(&mut self, insert: MergeableInsert) {
self.byte_len += insert.values.len() + 3;
self.values.push(insert.values);
self.source_sqls.push(insert.sql);
self.source_statement_count += 1;
}
fn to_sql(&self) -> String {
if self.source_statement_count == 1 {
return self.source_sqls.first().cloned().unwrap_or_default();
}
if self.syntax == InsertBatchSyntax::OracleInsertAll {
if let Some(target) = oracle_insert_all_target(&self.prefix) {
let mut sql = String::with_capacity(
self.byte_len + "INSERT ALL\n".len() + "SELECT 1 FROM dual".len() + target.len() + 14,
);
sql.push_str("INSERT ALL\n");
for values in &self.values {
sql.push_str("INTO ");
sql.push_str(target);
sql.push_str(" VALUES ");
sql.push_str(values);
sql.push('\n');
}
sql.push_str("SELECT 1 FROM dual");
return sql;
}
}
format!("{} VALUES\n{}", self.prefix, self.values.join(",\n"))
}
}
/// The `table (columns)` part of `INSERT INTO table (columns) VALUES …`, used
/// as the repeated `INTO` clause of an `INSERT ALL` batch.
fn oracle_insert_all_target(prefix: &str) -> Option<&str> {
let mut scanner = SqlScanner::default();
let mut seen_insert = false;
for (idx, ch) in prefix.char_indices() {
scanner.step(prefix, idx, ch);
if scanner.is_masked() {
continue;
}
if !seen_insert {
seen_insert = keyword_at(prefix, idx, "insert");
continue;
}
if keyword_at(prefix, idx, "into") {
let target = prefix[idx + "into".len()..].trim();
return (!target.is_empty()).then_some(target);
}
}
None
}
fn parse_mergeable_insert(sql: &str, options: SqlParsingOptions) -> Option<MergeableInsert> {
let executable = leading_executable_sql_with_options(sql, options).trim().trim_end_matches(';').trim();
if !starts_with_keyword(executable, "insert") {
return None;
}
let (values_start, values_end) = find_top_level_values_keyword(executable)?;
let prefix_without_values = executable[..values_start].trim_end();
let values = executable[values_end..].trim();
let values = parse_insert_values_tail(values)?;
let prefix = prefix_without_values.to_string();
Some(MergeableInsert {
prefix_key: normalize_insert_prefix_key(&prefix),
prefix,
values,
sql: executable.to_string(),
})
}
fn find_top_level_values_keyword(sql: &str) -> Option<(usize, usize)> {
let mut scanner = SqlScanner::default();
let mut depth = 0usize;
for (idx, ch) in sql.char_indices() {
scanner.step(sql, idx, ch);
if scanner.is_masked() {
continue;
}
match ch {
'(' => depth += 1,
')' => depth = depth.saturating_sub(1),
_ if depth == 0 => {
for keyword in ["values", "value"] {
if keyword_at(sql, idx, keyword) {
return Some((idx, idx + keyword.len()));
}
}
}
_ => {}
}
}
None
}
fn parse_insert_values_tail(tail: &str) -> Option<String> {
let tail = tail.trim().trim_end_matches(';').trim();
if tail.is_empty() {
return None;
}
let mut scanner = SqlScanner::default();
let mut depth = 0usize;
let mut saw_tuple = false;
let mut expecting_tuple = true;
for (idx, ch) in tail.char_indices() {
scanner.step(tail, idx, ch);
if scanner.is_masked() {
continue;
}
if expecting_tuple {
if ch.is_whitespace() || (saw_tuple && ch == ',') {
continue;
}
if ch != '(' {
return None;
}
expecting_tuple = false;
saw_tuple = true;
depth = 1;
continue;
}
match ch {
'(' => depth += 1,
')' => {
depth = depth.checked_sub(1)?;
if depth == 0 {
expecting_tuple = true;
}
}
',' if depth == 0 => {}
_ if depth == 0 && !ch.is_whitespace() => return None,
_ => {}
}
}
if saw_tuple && depth == 0 {
Some(tail.to_string())
} else {
None
}
}
#[derive(Default)]
struct SqlScanner {
profile: SqlDialectProfile,
in_single_quote: bool,
single_quote_escape_string: bool,
in_double_quote: bool,
in_backtick: bool,
in_line_comment: bool,
block_comment_depth: usize,
oracle_q_quote: OracleQQuoteState,
dollar_quote_tag: Option<String>,
previous: Option<char>,
}
impl SqlScanner {
fn with_profile(profile: SqlDialectProfile) -> Self {
Self { profile, ..Self::default() }
}
fn step(&mut self, sql: &str, idx: usize, ch: char) {
if let Some(tag) = self.dollar_quote_tag.clone() {
if sql[idx..].starts_with(&tag) {
self.dollar_quote_tag = None;
}
self.previous = Some(ch);
return;
}
let next = next_char_at(sql, idx + ch.len_utf8());
if self.in_line_comment {
if ch == '\n' {
self.in_line_comment = false;
}
self.previous = Some(ch);
return;
}
if self.block_comment_depth > 0 {
if self.profile.supports_nested_block_comments && self.previous == Some('/') && ch == '*' {
self.block_comment_depth += 1;
// A nested opener's asterisk must not pair with a following slash as a close.
self.previous = None;
return;
} else if self.previous == Some('*') && ch == '/' {
self.block_comment_depth -= 1;
}
self.previous = Some(ch);
return;
}
if self.oracle_q_quote.is_open() {
self.oracle_q_quote.consume(ch, self.previous);
self.previous = Some(ch);
return;
}
if !self.in_single_quote && !self.in_double_quote && !self.in_backtick {
if self.profile.supports_oracle_q_quotes && ch == '\'' && ends_with_oracle_q_quote_prefix(&sql[..idx]) {
self.oracle_q_quote.awaiting_delimiter = true;
self.previous = Some(ch);
return;
}
if (ch == '-' && next == Some('-')) || (self.profile.supports_hash_line_comments && ch == '#') {
self.in_line_comment = true;
} else if ch == '/' && next == Some('*') {
self.block_comment_depth = 1;
// The opener's own asterisk must not read as a nested opener.
self.previous = None;
return;
} else if let Some(tag) =
self.profile.supports_dollar_quoted_strings.then(|| dollar_quote_tag_at_str(sql, idx)).flatten()
{
self.dollar_quote_tag = Some(tag);
}
}
let single_quote_backslash_escapes = if self.in_single_quote {
self.single_quote_escape_string
} else {
self.profile.supports_backslash_escaped_quotes
};
match ch {
'\'' if !(self.in_double_quote
|| self.in_backtick
|| single_quote_backslash_escapes && self.previous == Some('\\')) =>
{
self.in_single_quote = !self.in_single_quote;
self.single_quote_escape_string = self.in_single_quote
&& (self.profile.supports_backslash_escaped_quotes
|| (self.profile.supports_postgres_escape_strings
&& ends_with_escape_string_prefix(&sql[..idx])));
}
'"' if !(self.in_single_quote
|| self.in_backtick
|| self.profile.supports_backslash_escaped_quotes && self.previous == Some('\\')) =>
{
self.in_double_quote = !self.in_double_quote;
}
'`' if self.profile.supports_backtick_identifiers && !self.in_single_quote && !self.in_double_quote => {
self.in_backtick = !self.in_backtick;
}
_ => {}
}
self.previous = Some(ch);
}
fn is_masked(&self) -> bool {
self.in_single_quote
|| self.in_double_quote
|| self.in_backtick
|| self.in_line_comment
|| self.block_comment_depth > 0
|| self.oracle_q_quote.is_open()
|| self.dollar_quote_tag.is_some()
}
}
fn keyword_at(sql: &str, idx: usize, keyword: &str) -> bool {
let end = idx + keyword.len();
sql.get(idx..end).is_some_and(|candidate| candidate.eq_ignore_ascii_case(keyword))
&& sql[..idx].chars().next_back().is_none_or(|ch| !is_sql_ident_char(ch))
&& sql.get(end..).and_then(|tail| tail.chars().next()).is_none_or(|ch| !is_sql_ident_char(ch))
}
fn starts_with_keyword(sql: &str, keyword: &str) -> bool {
sql.get(..keyword.len()).is_some_and(|candidate| candidate.eq_ignore_ascii_case(keyword))
&& sql.get(keyword.len()..).and_then(|tail| tail.chars().next()).is_none_or(|ch| !is_sql_ident_char(ch))
}
fn is_sql_ident_char(ch: char) -> bool {
ch == '_' || ch == '$' || ch.is_ascii_alphanumeric()
}
fn normalize_insert_prefix_key(prefix: &str) -> String {
let mut scanner = SqlScanner::default();
let mut key = String::with_capacity(prefix.len());
let mut previous_space = false;
for (idx, ch) in prefix.char_indices() {
scanner.step(prefix, idx, ch);
if scanner.in_single_quote
|| scanner.in_double_quote
|| scanner.in_backtick
|| scanner.dollar_quote_tag.is_some()
{
key.push(ch);
previous_space = false;
continue;
}
if ch.is_whitespace() {
if !previous_space {
key.push(' ');
}
previous_space = true;
} else {
key.push(ch.to_ascii_lowercase());
previous_space = false;
}
}
key.trim().to_string()
}
pub fn starts_with_executable_sql_keyword(sql: &str, keywords: &[&str]) -> bool {
starts_with_executable_sql_keyword_with_options(sql, keywords, SqlParsingOptions::default())
}
pub fn starts_with_executable_sql_keyword_for_database(sql: &str, keywords: &[&str], db_type: DatabaseType) -> bool {
starts_with_executable_sql_keyword_with_options(sql, keywords, SqlParsingOptions::for_database_type(db_type))
}
pub fn starts_with_duckdb_result_sql_keyword(sql: &str) -> bool {
starts_with_executable_sql_keyword(
sql,
&[
"SELECT",
"SHOW",
"DESCRIBE",
"EXPLAIN",
"WITH",
"PRAGMA",
"FROM",
"SUMMARIZE",
"SUMMARISE",
"PIVOT",
"UNPIVOT",
],
)
}
pub fn starts_with_executable_sql_keyword_with_options(
sql: &str,
keywords: &[&str],
options: SqlParsingOptions,
) -> bool {
let Some(token) = first_executable_sql_token_with_options(sql, options) else {
return false;
};
keywords.iter().any(|keyword| executable_sql_keyword_matches(token, keyword))
}
fn executable_sql_keyword_matches(token: &str, keyword: &str) -> bool {
token.eq_ignore_ascii_case(keyword)
|| (keyword.eq_ignore_ascii_case("DESCRIBE") && token.eq_ignore_ascii_case("DESC"))
}
fn is_mysql_compatible_import_profile(profile: &str) -> bool {
matches!(
profile,
"mariadb"
| "tidb"
| "oceanbase"
| "custom_mysql"
| "doris"
| "starrocks"
| "manticoresearch"
| "selectdb"
| "goldendb"
)
}
pub fn supports_connection_level_database_bootstrap_target(
db_type: &DatabaseType,
driver_profile: Option<&str>,
) -> bool {
matches!(db_type, DatabaseType::Mysql | DatabaseType::Doris | DatabaseType::StarRocks | DatabaseType::Goldendb)
|| driver_profile
.map(|profile| profile.to_ascii_lowercase())
.is_some_and(|profile| is_mysql_compatible_import_profile(&profile) && profile != "manticoresearch")
}
pub fn is_mysql_compatible_import_target(db_type: &DatabaseType, driver_profile: Option<&str>) -> bool {
matches!(
db_type,
DatabaseType::Mysql
| DatabaseType::Doris
| DatabaseType::StarRocks
| DatabaseType::ManticoreSearch
| DatabaseType::Goldendb
) || driver_profile
.map(|profile| profile.to_ascii_lowercase())
.is_some_and(|profile| is_mysql_compatible_import_profile(&profile))
}
fn mysql_executable_comment_body(statement: &str) -> Option<&str> {
let bytes = statement.as_bytes();
let start = leading_mysql_executable_comment_start(statement)?;
let body_start = if bytes.get(start + 2) == Some(&b'!') { start + 3 } else { start + 4 };
let mut body_start = body_start;
while body_start < bytes.len() && (bytes[body_start].is_ascii_digit() || bytes[body_start].is_ascii_whitespace()) {
body_start += 1;
}
let close = find_block_comment_close(bytes, body_start, false)?;
if has_executable_sql(&statement[close + 2..]) {
return None;
}
Some(&statement[body_start..close])
}
fn leading_mysql_executable_comment_start(statement: &str) -> Option<usize> {
let bytes = statement.as_bytes();
let mut i = 0;
while i < bytes.len() {
while i < bytes.len() && bytes[i].is_ascii_whitespace() {
i += 1;
}
if i + 1 < bytes.len() && bytes[i] == b'-' && bytes[i + 1] == b'-' {
i += 2;
while i < bytes.len() && bytes[i] != b'\n' {
i += 1;
}
continue;
}
if bytes[i] == b'#' {
i += 1;
while i < bytes.len() && bytes[i] != b'\n' {
i += 1;
}
continue;
}
if i + 1 < bytes.len() && bytes[i] == b'/' && bytes[i + 1] == b'*' {
if i + 2 < bytes.len() && (bytes[i + 2] == b'!' || (i + 3 < bytes.len() && &bytes[i + 2..i + 4] == b"M!")) {
return Some(i);
}
let close = find_block_comment_close(bytes, i + 2, false)?;
i = close + 2;
continue;
}
return None;
}
None
}
fn find_block_comment_close(bytes: &[u8], mut start: usize, nested: bool) -> Option<usize> {
let mut depth = 1usize;
while start + 1 < bytes.len() {
if bytes[start] == b'*' && bytes[start + 1] == b'/' {
depth -= 1;
if depth == 0 {
return Some(start);
}
start += 2;
continue;
}
if nested && bytes[start] == b'/' && bytes[start + 1] == b'*' {
depth += 1;
start += 2;
continue;
}
start += 1;
}
None
}
fn is_mysql_key_toggle_statement(statement: &str) -> bool {
let upper = statement.split_whitespace().collect::<Vec<_>>().join(" ").to_ascii_uppercase();
upper.starts_with("ALTER TABLE ") && (upper.ends_with(" ENABLE KEYS") || upper.ends_with(" DISABLE KEYS"))
}
fn is_mysql_lock_table_statement(statement: &str) -> bool {
let executable = leading_executable_sql(statement);
let upper = executable.split_whitespace().collect::<Vec<_>>().join(" ").to_ascii_uppercase();
upper == "UNLOCK TABLES" || (upper.starts_with("LOCK TABLES ") && upper.ends_with(" WRITE"))
}
fn is_mysql_session_restore_statement(statement: &str) -> bool {
let executable = leading_executable_sql_with_options(statement, SqlParsingOptions::mysql_compatible());
let upper = executable.split_whitespace().collect::<Vec<_>>().join(" ").to_ascii_uppercase();
if !upper.starts_with("SET ") {
return false;
}
let assignment = upper.trim_start_matches("SET ").trim();
if assignment.starts_with('@') {
return false;
}
assignment.contains("= @OLD_")
|| assignment.contains("=@OLD_")
|| assignment.contains("= @SAVED_")
|| assignment.contains("=@SAVED_")
}
fn leading_executable_sql(sql: &str) -> &str {
leading_executable_sql_with_options(sql, SqlParsingOptions::default())
}
fn leading_executable_sql_with_options(sql: &str, options: SqlParsingOptions) -> &str {
let bytes = sql.as_bytes();
let mut i = 0;
while i < bytes.len() {
while i < bytes.len() && bytes[i].is_ascii_whitespace() {
i += 1;
}
if i >= bytes.len() {
break;
}
if i + 1 < bytes.len() && bytes[i] == b'-' && bytes[i + 1] == b'-' {
i += 2;
while i < bytes.len() && bytes[i] != b'\n' {
i += 1;
}
continue;
}
if options.profile.supports_hash_line_comments && bytes[i] == b'#' {
i += 1;
while i < bytes.len() && bytes[i] != b'\n' {
i += 1;
}
continue;
}
if i + 1 < bytes.len() && bytes[i] == b'/' && bytes[i + 1] == b'*' {
if i + 2 < bytes.len() && (bytes[i + 2] == b'!' || (i + 3 < bytes.len() && &bytes[i + 2..i + 4] == b"M!")) {
break;
}
let Some(close) = find_block_comment_close(bytes, i + 2, options.profile.supports_nested_block_comments)
else {
return &sql[sql.len()..];
};
i = close + 2;
continue;
}
break;
}
&sql[i..]
}
fn first_executable_sql_token_with_options(sql: &str, options: SqlParsingOptions) -> Option<&str> {
let bytes = sql.as_bytes();
let mut i = 0;
while i < bytes.len() {
while i < bytes.len() && (bytes[i].is_ascii_whitespace() || bytes[i] == b'(') {
i += 1;
}
if i >= bytes.len() {
break;
}
if i + 1 < bytes.len() && bytes[i] == b'-' && bytes[i + 1] == b'-' {
i += 2;
while i < bytes.len() && bytes[i] != b'\n' {
i += 1;
}
continue;
}
if options.profile.supports_hash_line_comments && bytes[i] == b'#' {
i += 1;
while i < bytes.len() && bytes[i] != b'\n' {
i += 1;
}
continue;
}
if i + 1 < bytes.len() && bytes[i] == b'/' && bytes[i + 1] == b'*' {
if i + 2 < bytes.len() && (bytes[i + 2] == b'!' || (i + 3 < bytes.len() && &bytes[i + 2..i + 4] == b"M!")) {
i += if bytes[i + 2] == b'!' { 3 } else { 4 };
while i < bytes.len() && (bytes[i].is_ascii_digit() || bytes[i].is_ascii_whitespace()) {
i += 1;
}
break;
}
i += 2;
while i + 1 < bytes.len() && !(bytes[i] == b'*' && bytes[i + 1] == b'/') {
i += 1;
}
i = (i + 2).min(bytes.len());
continue;
}
break;
}
let start = i;
while i < bytes.len() && (bytes[i].is_ascii_alphabetic() || bytes[i] == b'_') {
i += 1;
}
(i > start).then_some(&sql[start..i])
}
fn starts_with_oracle_plsql_block(sql: &str) -> bool {
OraclePlSqlBlock::parse(sql).starts_block()
}
/// Whether a statement must stay a single statement instead of being cut at its inner
/// semicolons: Oracle PL/SQL blocks for Oracle-like dialects, and the Oracle-style routine bodies
/// that PostgreSQL-family connections reach on a GaussDB/openGauss Oracle-compatibility server.
fn keeps_oracle_style_block_together(options: SqlParsingOptions, statement_head: &str) -> bool {
if options.profile.supports_oracle_plsql_blocks {
return starts_with_oracle_plsql_block(statement_head);
}
options.profile.supports_oracle_style_routine_bodies && starts_with_oracle_style_routine_body(statement_head)
}
/// Whether the statement opened so far is a T-SQL control-flow batch that the
/// `;`-statement splitter must keep whole: an `IF`/`WHILE` batch (its block
/// body keeps the splitter going; a branchless one is a single statement
/// anyway), or a top-level `BEGIN` that opens a real block (`BEGIN TRY ... END
/// TRY BEGIN CATCH ... END CATCH` counts as one). `BEGIN TRAN[SACTION]` and
/// friends have no matching `END` and stay split per statement. Only meaningful
/// when the profile keeps such batches (`keeps_sqlserver_control_flow_batches`).
fn keeps_sqlserver_control_flow_batch(options: SqlParsingOptions, statement_head: &str) -> bool {
if !options.profile.keeps_sqlserver_control_flow_batches {
return false;
}
let tokens = sqlserver_top_level_tokens(statement_head);
let Some(first) = tokens.first() else {
return false;
};
if first.is_word("IF") || first.is_word("WHILE") {
return true;
}
first.is_word("BEGIN") && sqlserver_begins_block(&tokens, 0)
}
/// Top-level `BEGIN`/`CASE` blocks the T-SQL statement has left open at its
/// end. `END TRY` immediately followed by `BEGIN CATCH` continues the TRY
/// block, so the pair keeps the depth open instead of ending the batch
/// (#13051). Tokens inside strings, bracketed identifiers, comments and
/// parentheses are ignored, so a nested statement's own blocks do not matter.
fn sqlserver_control_flow_open_blocks(sql: &str) -> usize {
let tokens = sqlserver_top_level_tokens(sql);
let mut depth = 0usize;
let mut carried_try = false;
for (index, token) in tokens.iter().enumerate() {
if token.is_word("BEGIN") {
if !sqlserver_begins_block(&tokens, index) {
continue;
}
if carried_try {
// `BEGIN CATCH` continues the carried TRY block: the depth the
// `END TRY` left open is consumed here, not extended.
carried_try = false;
} else {
depth += 1;
}
continue;
}
if token.is_word("CASE") {
depth += 1;
continue;
}
if !token.is_word("END") {
continue;
}
let next = tokens.get(index + 1);
// `END CONVERSATION`/`END DIALOG` close a conversation, not a block.
if next.is_some_and(|next| next.is_word("CONVERSATION") || next.is_word("DIALOG")) {
continue;
}
if next.is_some_and(|next| next.is_word("TRY"))
&& tokens.get(index + 2).is_some_and(|token| token.is_word("BEGIN"))
&& tokens.get(index + 3).is_some_and(|token| token.is_word("CATCH"))
{
// `END TRY` immediately followed by `BEGIN CATCH`: both halves are
// one batch, so the CATCH block continues the TRY block's span.
carried_try = true;
continue;
}
depth = depth.saturating_sub(1);
carried_try = false;
}
depth
}
/// Whether the buffered control-flow batch is complete: every top-level
/// `BEGIN`/`CASE` it opened has been closed, so this `;` is a real statement
/// boundary. A batch still open — including one that is mid-typing — keeps the
/// `;` and stays whole.
fn sqlserver_control_flow_batch_is_complete(sql: &str) -> bool {
sqlserver_control_flow_open_blocks(sql) == 0
}
/// Byte offset just after the `END`/`END TRY`/`END CATCH` that closes the block
/// a `;`-fragment carried into it, when that closure is final (not followed by
/// `ELSE`). T-SQL semicolons are optional, so a batch whose final `END` has no
/// trailing `;` shares its fragment with the next statement; the statement
/// boundary then sits right after the closing token instead of at the next
/// `;`. Mirrors `scanSqlServerControlFlow`'s `closedAt`/`wordAfterClose` in
/// `sqlStatementRanges.ts`: a block opened and closed inside one `;`-free run
/// has no boundary (the whole run stays one statement), and a closure followed
/// by `ELSE` only continues the same `IF`, so a later closure in the same
/// fragment replaces it.
fn sqlserver_control_flow_batch_boundary(sql: &str) -> Option<usize> {
let tokens = sqlserver_top_level_tokens(sql);
let mut index = 0usize;
let mut fragment_start_depth = 0usize;
while index < tokens.len() {
// Scan one `;`-delimited fragment with the depth the previous fragment
// left open; only a closure of that carried depth is a boundary.
let mut depth = fragment_start_depth;
let mut carried_try = false;
let mut boundary: Option<(usize, &str)> = None;
while index < tokens.len() {
let token = &tokens[index];
if token.is_word(";") {
index += 1;
break;
}
if token.is_word("BEGIN") {
if sqlserver_begins_block(&tokens, index) {
if carried_try {
// `BEGIN CATCH` continues the carried TRY block.
carried_try = false;
} else {
depth += 1;
}
}
index += 1;
continue;
}
if token.is_word("CASE") {
depth += 1;
index += 1;
continue;
}
if !token.is_word("END") {
index += 1;
continue;
}
let next = tokens.get(index + 1);
// `END CONVERSATION`/`END DIALOG` close a conversation, not a block.
if next.is_some_and(|next| next.is_word("CONVERSATION") || next.is_word("DIALOG")) {
index += 1;
continue;
}
if next.is_some_and(|next| next.is_word("TRY"))
&& tokens.get(index + 2).is_some_and(|token| token.is_word("BEGIN"))
&& tokens.get(index + 3).is_some_and(|token| token.is_word("CATCH"))
{
// `END TRY BEGIN CATCH` keeps the pair in one batch.
carried_try = true;
index += 1;
continue;
}
depth = depth.saturating_sub(1);
carried_try = false;
if fragment_start_depth > 0 && depth == 0 {
// `END TRY`/`END CATCH` are two-token closers: the boundary sits
// after the pair, and the word after the pair decides finality.
let paired = next.is_some_and(|next| next.is_word("TRY") || next.is_word("CATCH"));
let (end_offset, word_after) = if paired {
(tokens[index + 1].end, tokens.get(index + 2).map_or("", |token| token.text))
} else {
(token.end, next.map_or("", |token| token.text))
};
if boundary.as_ref().is_none_or(|(_, word)| word.eq_ignore_ascii_case("ELSE")) {
boundary = Some((end_offset, word_after));
}
}
index += 1;
}
if let Some((offset, word_after)) = boundary {
if !word_after.eq_ignore_ascii_case("ELSE") {
return Some(offset);
}
}
fragment_start_depth = depth;
}
None
}
fn starts_with_oracle_style_routine_body(sql: &str) -> bool {
OraclePlSqlBlock::parse(sql).starts_oracle_style_routine_body()
}
fn starts_with_postgres_dollar_quoted_routine_prefix(sql: &str) -> bool {
let block = OraclePlSqlBlock::parse(sql);
let Some(first) = block.tokens.first() else {
return false;
};
if !first.is_word("CREATE") {
return false;
}
let tokens = OraclePlSqlBlock::skip_create_modifiers(&block.tokens[1..]);
tokens.first().is_some_and(|token| token.is_any_word(&["FUNCTION", "PROCEDURE"]))
&& block.tokens.iter().rev().find_map(OraclePlSqlToken::as_word) == Some("AS")
}
pub(crate) fn oracle_plsql_block_is_complete(sql: &str) -> bool {
OraclePlSqlBlock::parse(sql).is_complete()
}
fn starts_with_hana_script_block(sql: &str) -> bool {
HanaScriptBlock::parse(sql).starts_block()
}
fn hana_script_block_is_complete(sql: &str) -> bool {
if let Some((_, last_line)) = sql.rsplit_once('\n') {
if !last_line.trim_start().to_ascii_uppercase().starts_with("END;") {
return false;
}
}
HanaScriptBlock::parse(sql).is_complete()
}
struct HanaScriptBlock {
tokens: Vec<OraclePlSqlToken>,
}
impl HanaScriptBlock {
fn parse(sql: &str) -> Self {
Self { tokens: oracle_plsql_tokens(sql) }
}
fn starts_block(&self) -> bool {
let Some(first) = self.tokens.first().and_then(OraclePlSqlToken::as_word) else {
return false;
};
if first == "DO" {
return true;
}
let mut index = 1;
if first == "CREATE"
&& self.tokens.get(index).is_some_and(|token| token.is_word("OR"))
&& self.tokens.get(index + 1).is_some_and(|token| token.is_word("REPLACE"))
{
index += 2;
}
let Some(object_type) = self.tokens.get(index).and_then(OraclePlSqlToken::as_word) else {
return false;
};
if first != "CREATE" && first != "ALTER" {
return false;
}
if !matches!(object_type, "PROCEDURE" | "FUNCTION" | "TRIGGER") {
return false;
}
// HANA also accepts ALTER PROCEDURE ... RECOMPILE. Only hold a statement
// as a script block once its SQLScript body is present, otherwise a plain
// metadata/compilation ALTER would never flush at its semicolon.
self.tokens.iter().any(|token| token.is_word("BEGIN"))
}
fn is_complete(&self) -> bool {
if !self.starts_block() {
return false;
}
let mut block_depth = 0usize;
let mut case_depth = 0usize;
let mut saw_begin = false;
for (index, token) in self.tokens.iter().enumerate() {
if token.is_word("BEGIN") {
if previous_adjacent_word(&self.tokens, index).is_some_and(|previous| previous == "END") {
continue;
}
block_depth += 1;
saw_begin = true;
continue;
}
if token.is_word("CASE") {
if previous_adjacent_word(&self.tokens, index).is_none_or(|previous| previous != "END") {
case_depth += 1;
}
continue;
}
if token.is_word("END") {
let next = next_adjacent_word(&self.tokens, index);
if matches!(next, Some("IF" | "FOR" | "WHILE")) {
continue;
}
if case_depth > 0 {
case_depth -= 1;
} else {
block_depth = block_depth.saturating_sub(1);
}
}
}
saw_begin
&& block_depth == 0
&& case_depth == 0
&& self.tokens.last().is_some_and(OraclePlSqlToken::is_semicolon)
}
}
fn starts_with_firebird_routine_block(sql: &str) -> bool {
FirebirdRoutineBlock::parse(sql).starts_block()
}
fn firebird_routine_block_is_complete(sql: &str) -> bool {
FirebirdRoutineBlock::parse(sql).is_complete()
}
struct FirebirdRoutineBlock {
tokens: Vec<OraclePlSqlToken>,
}
impl FirebirdRoutineBlock {
fn parse(sql: &str) -> Self {
Self { tokens: oracle_plsql_tokens(sql) }
}
fn starts_block(&self) -> bool {
let Some(first) = self.tokens.first().and_then(OraclePlSqlToken::as_word) else {
return false;
};
if first == "EXECUTE" {
return self.tokens.get(1).is_some_and(|token| token.is_word("BLOCK"));
}
if !matches!(first, "CREATE" | "RECREATE" | "ALTER") {
return false;
}
let mut index = 1;
if self.tokens.get(index).is_some_and(|token| token.is_word("OR"))
&& self.tokens.get(index + 1).is_some_and(|token| token.is_any_word(&["ALTER", "REPLACE"]))
{
index += 2;
}
let Some(object_type) = self.tokens.get(index).and_then(OraclePlSqlToken::as_word) else {
return false;
};
if object_type == "PACKAGE" {
if !self.tokens.get(index + 1).is_some_and(|token| token.is_word("BODY")) {
return false;
}
} else if !matches!(object_type, "PROCEDURE" | "FUNCTION" | "TRIGGER") {
return false;
}
if first == "ALTER" {
// Firebird accepts `ALTER TRIGGER name ACTIVE/INACTIVE;`. Only treat ALTER
// as a routine block when it defines a body via AS or BEGIN.
return self.tokens.iter().any(|token| token.is_any_word(&["AS", "BEGIN"]));
}
true
}
fn is_complete(&self) -> bool {
if !self.starts_block() {
return false;
}
let mut block_depth = 0usize;
let mut case_depth = 0usize;
let mut saw_begin = false;
for (index, token) in self.tokens.iter().enumerate() {
if token.is_word("BEGIN") {
if previous_adjacent_word(&self.tokens, index).is_some_and(|previous| previous == "END") {
continue;
}
block_depth += 1;
saw_begin = true;
continue;
}
if token.is_word("CASE") {
if previous_adjacent_word(&self.tokens, index).is_none_or(|previous| previous != "END") {
case_depth += 1;
}
continue;
}
if token.is_word("END") {
if case_depth > 0 {
case_depth -= 1;
} else {
block_depth = block_depth.saturating_sub(1);
}
}
}
saw_begin
&& block_depth == 0
&& case_depth == 0
&& self.tokens.last().is_some_and(OraclePlSqlToken::is_semicolon)
}
}
/// The BEGIN/END scope walk behind [`OraclePlSqlBlock::is_complete`], kept resumable so
/// [`OraclePlSqlBlockTracker`] only walks the tokens added since its previous check.
#[derive(Debug, Clone)]
struct OraclePlSqlBlockWalk {
object_kind: Option<OraclePlSqlCreateObjectKind>,
scopes: Vec<OraclePlSqlScope>,
awaiting_routine_declarations: bool,
saw_begin: bool,
complete: bool,
next_index: usize,
}
impl OraclePlSqlBlockWalk {
fn new(block: &OraclePlSqlBlock) -> Self {
// Package/type specifications have declarations and an outer END with
// no BEGIN. Bodies also own an outer END beyond nested routine END
// pairs so an inner END cannot finish the object.
let object_kind = block.create_object_kind();
let scopes = match object_kind {
Some(OraclePlSqlCreateObjectKind::Body | OraclePlSqlCreateObjectKind::Spec) => {
vec![OraclePlSqlScope::Object]
}
None => Vec::new(),
};
Self {
object_kind,
scopes,
// A stored PROCEDURE/FUNCTION has no DECLARE keyword: its declare section
// starts right after its own IS/AS. Opening it there lets a local subprogram
// own its BEGIN ... END instead of ending the routine.
awaiting_routine_declarations: block.is_create_routine(),
saw_begin: false,
complete: false,
next_index: 0,
}
}
fn advance(&mut self, tokens: &[OraclePlSqlToken]) {
while self.next_index < tokens.len() {
self.step(tokens, self.next_index);
self.next_index += 1;
}
}
fn step(&mut self, tokens: &[OraclePlSqlToken], index: usize) {
let token = &tokens[index];
let scopes = &mut self.scopes;
if token.is_semicolon() {
if matches!(scopes.last(), Some(OraclePlSqlScope::RoutineHeader)) {
scopes.pop();
}
return;
}
if self.awaiting_routine_declarations && token.is_any_word(&["IS", "AS"]) {
self.awaiting_routine_declarations = false;
scopes.push(OraclePlSqlScope::Declaration);
} else if token.is_word("DECLARE") {
if !matches!(scopes.last(), Some(OraclePlSqlScope::Declaration)) {
scopes.push(OraclePlSqlScope::Declaration);
}
} else if token.is_any_word(&["PROCEDURE", "FUNCTION"])
&& matches!(scopes.last(), Some(OraclePlSqlScope::Declaration | OraclePlSqlScope::Routine))
{
scopes.push(OraclePlSqlScope::RoutineHeader);
} else if token.is_any_word(&["IS", "AS"]) && matches!(scopes.last(), Some(OraclePlSqlScope::RoutineHeader)) {
*scopes.last_mut().unwrap() = OraclePlSqlScope::Routine;
} else if token.is_word("BEGIN") {
// A local routine's BEGIN belongs to that routine, not to the outer block.
if matches!(scopes.last(), Some(OraclePlSqlScope::Declaration | OraclePlSqlScope::Routine)) {
*scopes.last_mut().unwrap() = OraclePlSqlScope::Block;
} else {
scopes.push(OraclePlSqlScope::Block);
}
self.saw_begin = true;
self.complete = false;
} else if token.is_word("CASE") {
// Both CASE expressions and CASE statements own an END.
// The CASE token that follows END CASE is a suffix, not a scope start.
// Require source adjacency: `end ), (case` must open a new CASE scope.
if previous_adjacent_word(tokens, index) != Some("END") {
scopes.push(OraclePlSqlScope::Case);
}
} else if token.is_word("END") {
let next = next_adjacent_word(tokens, index);
if matches!(next, Some("IF" | "LOOP")) {
return;
}
if matches!(next, Some("CASE")) && !matches!(scopes.last(), Some(OraclePlSqlScope::Case)) {
return;
}
if scopes.pop().is_some() {
self.complete = scopes.is_empty();
}
}
}
fn is_complete(&self) -> bool {
match self.object_kind {
// Package/type objects complete on their outer END [name]; a package
// body may omit an initialization BEGIN when it has no executable
// initialization section.
Some(OraclePlSqlCreateObjectKind::Body | OraclePlSqlCreateObjectKind::Spec) => self.complete,
_ => self.saw_begin && self.complete,
}
}
}
/// Incremental [`keeps_oracle_style_block_together`] and [`oracle_plsql_block_is_complete`]
/// for a statement that the splitters grow one statement-level `;` at a time. Re-tokenizing
/// the whole statement at every inner `;` made splitting a large package body quadratic: a
/// 370 KB DM8 package body spent minutes in the splitter before reaching the server (#13450).
///
/// The splitters only consult it at a `;` outside quotes and comments, so tokenizing the
/// statement in slices that start or end at those semicolons yields the same tokens as
/// tokenizing it whole.
#[derive(Debug, Default)]
struct OraclePlSqlBlockTracker {
block: OraclePlSqlBlock,
scanned_len: usize,
walk: Option<OraclePlSqlBlockWalk>,
}
impl OraclePlSqlBlockTracker {
fn reset(&mut self) {
*self = Self::default();
}
fn scan(&mut self, statement: &str) {
if statement.len() < self.scanned_len {
self.reset();
}
self.block.tokens.extend(oracle_plsql_tokens(&statement[self.scanned_len..]));
self.scanned_len = statement.len();
}
/// [`keeps_oracle_style_block_together`] for `statement_head`, the statement before the `;`.
fn keeps_block_together(&mut self, options: SqlParsingOptions, statement_head: &str) -> bool {
if !options.profile.supports_oracle_plsql_blocks && !options.profile.supports_oracle_style_routine_bodies {
return false;
}
self.scan(statement_head);
if options.profile.supports_oracle_plsql_blocks {
return self.block.starts_block();
}
self.block.starts_oracle_style_routine_body()
}
/// [`oracle_plsql_block_is_complete`] for `statement`, which ends with the `;`.
fn is_complete(&mut self, statement: &str) -> bool {
self.scan(statement);
if let Some(complete) = self.block.completion_without_block_walk() {
return complete;
}
let block = &self.block;
let walk = self.walk.get_or_insert_with(|| OraclePlSqlBlockWalk::new(block));
walk.advance(&block.tokens);
walk.is_complete()
}
}
#[derive(Debug, Default)]
struct OraclePlSqlBlock {
tokens: Vec<OraclePlSqlToken>,
}
#[derive(Debug, Clone, PartialEq, Eq)]
enum OraclePlSqlToken {
Word(String),
/// A double-quoted identifier (`"schema"."procedure"`). Kept as its own variant
/// instead of a [`Self::Word`]: the block detectors compare keywords, and a quoted
/// name must never satisfy `is_word` (a column called `"END"` is not the block end).
/// Dropping it entirely made `BEGIN "S"."P"(); END;` tokenize exactly like a
/// transaction `BEGIN;`, so the splitter cut the block at the inner semicolon and
/// the server rejected the truncated statement (#10434).
QuotedIdentifier,
Semicolon,
/// Punctuation / operators / numbers. Kept so `end ), (case` is not treated as the
/// PL/SQL closer `END CASE` (adjacent-word checks must not skip across these).
Other,
}
impl OraclePlSqlBlock {
fn parse(sql: &str) -> Self {
Self { tokens: oracle_plsql_tokens(sql) }
}
fn starts_block(&self) -> bool {
match self.tokens.as_slice() {
[first, ..] if first.is_word("DECLARE") => true,
[first, second, ..] if first.is_word("BEGIN") && !Self::is_transaction_begin_tail(second) => true,
// Oracle 12c `WITH [FUNCTION|PROCEDURE ... END;] <main query>` inline PL/SQL
// declarations (also accepted by DM8 — issue #11275). The declaration section is
// PL/SQL, but the statement is not over at its `END;` — the main query's own
// terminator ends it.
[first, second, ..] if first.is_word("WITH") && second.is_any_word(&["FUNCTION", "PROCEDURE"]) => true,
[first, rest @ ..] if first.is_word("CREATE") => Self::starts_create_plsql_object(rest),
_ => false,
}
}
/// Whether the statement is a routine whose body is written in Oracle PL/SQL syntax
/// (`CREATE [OR REPLACE] PROCEDURE|FUNCTION ... { AS | IS } { DECLARE | BEGIN }`, plus the
/// Oracle-only `PACKAGE BODY` / `TYPE BODY` forms).
///
/// This is deliberately narrower than [`Self::starts_block`]. A PostgreSQL connection only
/// sees this shape when the server is a GaussDB/openGauss instance in Oracle (A) compatibility
/// mode, because PostgreSQL itself requires a dollar-quoted or string body
/// (`AS $$ ... $$`, `AS 'body'`) and has no `PACKAGE`/`TYPE BODY`. Keeping the narrow test at
/// the entry point lets [`Self::is_complete`] end such a body at its outer `END;` while
/// ordinary PostgreSQL routines, `CREATE TRIGGER ... EXECUTE FUNCTION` statements and
/// transaction `BEGIN;` blocks keep splitting at their own statement terminator.
fn starts_oracle_style_routine_body(&self) -> bool {
let [first, rest @ ..] = self.tokens.as_slice() else {
return false;
};
if !first.is_word("CREATE") {
return false;
}
let rest = Self::skip_create_modifiers(rest);
match rest {
[object, body, ..] if object.is_any_word(&["PACKAGE", "TYPE"]) && body.is_word("BODY") => true,
[object, ..] if object.is_any_word(&["FUNCTION", "PROCEDURE"]) => {
Self::routine_body_starts_with_plsql_keyword(rest)
}
_ => false,
}
}
/// The body introducer of an Oracle-style routine. Declarations may follow `AS`/`IS` either
/// directly or after the optional `DECLARE` keyword, so only the `DECLARE`/`BEGIN` forms are
/// recognized; a dollar-quoted or string body is never a word token, so PostgreSQL routines
/// cannot match here.
fn routine_body_starts_with_plsql_keyword(tokens: &[OraclePlSqlToken]) -> bool {
for (index, token) in tokens.iter().enumerate() {
if !token.is_any_word(&["AS", "IS"]) {
continue;
}
return tokens.get(index + 1).is_some_and(|next| next.is_any_word(&["DECLARE", "BEGIN"]));
}
false
}
fn is_complete(&self) -> bool {
if let Some(complete) = self.completion_without_block_walk() {
return complete;
}
let mut walk = OraclePlSqlBlockWalk::new(self);
walk.advance(&self.tokens);
walk.is_complete()
}
/// The completion rules decided by the statement header alone; `None` means the
/// BEGIN/END walk in [`OraclePlSqlBlockWalk`] decides.
fn completion_without_block_walk(&self) -> Option<bool> {
if !self.starts_block() {
return Some(false);
}
// `WITH [FUNCTION|PROCEDURE ... END;] <main query>;` follows its own completion
// rule (see with_function_is_complete); the classic block walk would fire at the
// declaration's `END;` and cut the main query off the statement.
if self.starts_with_function_declaration() {
return Some(self.with_function_is_complete());
}
if self.tokens.first().is_some_and(|token| token.is_word("CREATE")) {
let object = Self::skip_create_modifiers(&self.tokens[1..]);
if object.first().is_some_and(|token| token.is_word("TRIGGER"))
&& object
.iter()
.find(|token| token.is_any_word(&["BEGIN", "DECLARE", "COMPOUND", "CALL"]))
.is_some_and(|token| token.is_word("CALL"))
{
return Some(self.tokens.last().is_some_and(OraclePlSqlToken::is_semicolon));
}
}
None
}
fn starts_with_function_declaration(&self) -> bool {
matches!(
self.tokens.as_slice(),
[first, second, ..] if first.is_word("WITH") && second.is_any_word(&["FUNCTION", "PROCEDURE"])
)
}
/// Completion rule for `WITH [FUNCTION|PROCEDURE ... END;] <main query>;` (issue #11275):
/// no depth-0 `;` may end the statement while the WITH declaration section is open —
/// declaration assignments, nested block ends and the `END;` closing each declaration all
/// stay inside. Once the main query starts at an empty scope, its own terminator
/// completes the statement.
fn with_function_is_complete(&self) -> bool {
if !self.tokens.last().is_some_and(OraclePlSqlToken::is_semicolon) {
return false;
}
let mut scopes: Vec<OraclePlSqlScope> = Vec::new();
let mut in_main_query = false;
for (index, token) in self.tokens.iter().enumerate() {
if token.is_semicolon() {
if in_main_query && scopes.is_empty() {
return true;
}
continue;
}
let Some(word) = token.as_word() else {
continue;
};
// The statement-initial WITH token sits at index 0 and must not open the query.
if index > 0
&& scopes.is_empty()
&& matches!(word, "SELECT" | "INSERT" | "UPDATE" | "DELETE" | "MERGE" | "WITH")
{
in_main_query = true;
continue;
}
match word {
"BEGIN" => scopes.push(OraclePlSqlScope::Block),
"CASE" if previous_adjacent_word(&self.tokens, index) != Some("END") => {
scopes.push(OraclePlSqlScope::Case);
}
"END" => {
let next = next_adjacent_word(&self.tokens, index);
if matches!(next, Some("IF" | "LOOP")) {
continue;
}
if matches!(next, Some("CASE")) && !matches!(scopes.last(), Some(OraclePlSqlScope::Case)) {
continue;
}
scopes.pop();
}
_ => {}
}
}
false
}
fn starts_create_plsql_object(tokens: &[OraclePlSqlToken]) -> bool {
let tokens = Self::skip_create_modifiers(tokens);
match tokens {
// PACKAGE BODY / TYPE BODY are programmable blocks with an outer END.
[object, body, ..] if object.is_any_word(&["PACKAGE", "TYPE"]) && body.is_word("BODY") => true,
// Plain CREATE TYPE ... AS OBJECT (...); ends with ");" — not a PL/SQL block.
[object, ..] if object.is_word("TYPE") => false,
[object, ..] if object.is_any_word(&["FUNCTION", "PROCEDURE", "TRIGGER", "PACKAGE"]) => true,
_ => false,
}
}
fn create_object_kind(&self) -> Option<OraclePlSqlCreateObjectKind> {
if self.tokens.first().is_none_or(|token| !token.is_word("CREATE")) {
return None;
}
let tokens = Self::skip_create_modifiers(&self.tokens[1..]);
match tokens {
[object, body, ..] if object.is_any_word(&["PACKAGE", "TYPE"]) && body.is_word("BODY") => {
Some(OraclePlSqlCreateObjectKind::Body)
}
// A compound trigger's timing-point sections each own a BEGIN ... END
// <timing point>, and only the outer END [trigger] finishes it.
[object, header @ ..] if object.is_word("TRIGGER") && Self::is_compound_trigger_header(header) => {
Some(OraclePlSqlCreateObjectKind::Body)
}
// Only PACKAGE specs lack BEGIN; plain TYPE objects are ordinary SQL.
[object, ..] if object.is_word("PACKAGE") => Some(OraclePlSqlCreateObjectKind::Spec),
_ => None,
}
}
/// Whether a trigger header (the tokens after `TRIGGER`) declares a `COMPOUND TRIGGER`.
/// Only the header is inspected: it ends at the body of a simple trigger.
fn is_compound_trigger_header(tokens: &[OraclePlSqlToken]) -> bool {
let header_end = tokens
.iter()
.position(|token| token.is_semicolon() || token.is_any_word(&["BEGIN", "DECLARE", "CALL"]))
.unwrap_or(tokens.len());
tokens[..header_end].windows(2).any(|pair| pair[0].is_word("COMPOUND") && pair[1].is_word("TRIGGER"))
}
/// `CREATE [OR REPLACE] [EDITIONABLE] PROCEDURE|FUNCTION ...`.
fn is_create_routine(&self) -> bool {
let [first, rest @ ..] = self.tokens.as_slice() else {
return false;
};
first.is_word("CREATE")
&& Self::skip_create_modifiers(rest)
.first()
.is_some_and(|object| object.is_any_word(&["PROCEDURE", "FUNCTION"]))
}
/// Skip OR REPLACE / FORCE / NOFORCE / EDITIONABLE modifiers after CREATE.
fn skip_create_modifiers(tokens: &[OraclePlSqlToken]) -> &[OraclePlSqlToken] {
let mut rest = tokens;
loop {
match rest {
[or, replace, tail @ ..] if or.is_word("OR") && replace.is_word("REPLACE") => {
rest = tail;
}
[modifier, tail @ ..]
if modifier.is_any_word(&["FORCE", "NOFORCE", "EDITIONABLE", "NONEDITIONABLE"]) =>
{
rest = tail;
}
_ => break,
}
}
rest
}
fn is_transaction_begin_tail(token: &OraclePlSqlToken) -> bool {
token.is_semicolon() || token.is_any_word(&["TRANSACTION", "WORK"])
}
}
#[derive(Debug, Clone, Copy, PartialEq, Eq)]
enum OraclePlSqlCreateObjectKind {
Spec,
Body,
}
#[derive(Debug, Clone, Copy, PartialEq, Eq)]
enum OraclePlSqlScope {
Object,
/// DECLARE section awaiting its own BEGIN.
Declaration,
/// Local PROCEDURE/FUNCTION header, before IS/AS (a `;` here is a forward declaration).
RoutineHeader,
/// Local PROCEDURE/FUNCTION after IS/AS, awaiting its own BEGIN.
Routine,
Block,
Case,
}
impl OraclePlSqlToken {
fn word(value: String) -> Self {
Self::Word(value)
}
fn from_sqlparser_token(token: Token) -> Option<Self> {
match token {
Token::Word(word) if word.quote_style.is_none() => Some(Self::Word(word.value.to_ascii_uppercase())),
Token::Word(_) => Some(Self::QuotedIdentifier),
Token::SemiColon => Some(Self::Semicolon),
Token::Whitespace(_) | Token::Placeholder(_) => None,
_ => Some(Self::Other),
}
}
fn is_word(&self, expected: &str) -> bool {
matches!(self, Self::Word(value) if value == expected)
}
fn as_word(&self) -> Option<&str> {
match self {
Self::Word(value) => Some(value),
Self::QuotedIdentifier | Self::Semicolon | Self::Other => None,
}
}
fn is_any_word(&self, expected: &[&str]) -> bool {
expected.iter().any(|word| self.is_word(word))
}
fn is_semicolon(&self) -> bool {
matches!(self, Self::Semicolon)
}
}
/// Word immediately before `index` in the token stream (no skipping punctuation).
/// Used for `END CASE` / `END IF` pairing so `end ), (case` is not treated as `END CASE`.
fn previous_adjacent_word(tokens: &[OraclePlSqlToken], index: usize) -> Option<&str> {
index.checked_sub(1).and_then(|previous| tokens.get(previous)).and_then(OraclePlSqlToken::as_word)
}
fn next_adjacent_word(tokens: &[OraclePlSqlToken], index: usize) -> Option<&str> {
tokens.get(index + 1).and_then(OraclePlSqlToken::as_word)
}
fn push_other_token(tokens: &mut Vec<OraclePlSqlToken>) {
if tokens.last() != Some(&OraclePlSqlToken::Other) {
tokens.push(OraclePlSqlToken::Other);
}
}
fn oracle_plsql_tokens(sql: &str) -> Vec<OraclePlSqlToken> {
let dialect = OracleDialect {};
if let Ok(tokens) = Tokenizer::new(&dialect, sql).tokenize() {
let mut out = Vec::new();
for token in tokens {
if let Some(mapped) = OraclePlSqlToken::from_sqlparser_token(token) {
if mapped == OraclePlSqlToken::Other {
push_other_token(&mut out);
} else {
out.push(mapped);
}
}
}
return out;
}
oracle_plsql_tokens_fallback(sql)
}
fn oracle_plsql_tokens_fallback(sql: &str) -> Vec<OraclePlSqlToken> {
let mut tokens = Vec::new();
let mut iter = sql.char_indices().peekable();
while let Some((_, ch)) = iter.next() {
if ch.is_whitespace() {
continue;
}
if ch == '-' && iter.peek().is_some_and(|(_, next)| *next == '-') {
iter.next();
for (_, comment_ch) in iter.by_ref() {
if comment_ch == '\n' {
break;
}
}
continue;
}
if ch == '/' && iter.peek().is_some_and(|(_, next)| *next == '*') {
iter.next();
let mut previous = '\0';
for (_, comment_ch) in iter.by_ref() {
if previous == '*' && comment_ch == '/' {
break;
}
previous = comment_ch;
}
continue;
}
if ch == '\'' {
while let Some((_, quote_ch)) = iter.next() {
if quote_ch == '\'' {
if iter.peek().is_some_and(|(_, next)| *next == '\'') {
iter.next();
} else {
break;
}
}
}
// String literals separate keywords the same way punctuation does.
push_other_token(&mut tokens);
continue;
}
if ch == '"' {
for (_, ident_ch) in iter.by_ref() {
if ident_ch == '"' {
break;
}
}
tokens.push(OraclePlSqlToken::QuotedIdentifier);
continue;
}
if ch == ';' {
tokens.push(OraclePlSqlToken::Semicolon);
continue;
}
if ch.is_ascii_alphabetic() || ch == '_' {
let mut token = String::new();
token.push(ch.to_ascii_uppercase());
while let Some((_, next)) = iter.peek().copied() {
if next.is_ascii_alphanumeric() || next == '_' || next == '$' || next == '#' {
token.push(next.to_ascii_uppercase());
iter.next();
} else {
break;
}
}
tokens.push(OraclePlSqlToken::word(token));
continue;
}
push_other_token(&mut tokens);
}
tokens
}
fn starts_with_chars(chars: &[char], start: usize, needle: &[char]) -> bool {
start + needle.len() <= chars.len() && chars[start..start + needle.len()] == *needle
}
fn dollar_quote_tag_at(chars: &[char], start: usize) -> Option<String> {
if chars.get(start) != Some(&'$') {
return None;
}
match chars.get(start + 1) {
Some('$') => return Some("$$".to_string()),
Some(ch) if ch.is_ascii_alphabetic() || *ch == '_' => {}
_ => return None,
}
let mut end = start + 2;
while let Some(ch) = chars.get(end) {
if *ch == '$' {
return Some(chars[start..=end].iter().collect());
}
if !ch.is_ascii_alphanumeric() && *ch != '_' {
return None;
}
end += 1;
}
None
}
pub fn has_executable_sql(statement: &str) -> bool {
has_executable_sql_with_options(statement, SqlParsingOptions::default())
}
pub fn has_executable_sql_for_database(statement: &str, db_type: DatabaseType) -> bool {
has_executable_sql_with_options(statement, SqlParsingOptions::for_database_type(db_type))
}
fn executable_sql_bounds(statement: &str, options: SqlParsingOptions) -> Option<(usize, usize)> {
let trimmed_end = statement.trim_end().len();
let trimmed = &statement[..trimmed_end];
let executable = leading_executable_sql_with_options(trimmed, options);
if executable.is_empty() {
return None;
}
let executable_start = trimmed.len() - executable.len();
let start = if options.profile.preserves_tdsql_leading_directives {
crate::tdsql_mysql::leading_directive_start(trimmed, executable_start).unwrap_or(executable_start)
} else {
executable_start
};
Some((start, trimmed_end))
}
fn has_executable_sql_with_options(statement: &str, options: SqlParsingOptions) -> bool {
let chars = statement.chars().collect::<Vec<_>>();
let mut in_line_comment = false;
let mut block_comment_depth = 0usize;
let mut previous = None;
let mut i = 0;
while i < chars.len() {
let ch = chars[i];
let next = chars.get(i + 1).copied();
if in_line_comment {
if ch == '\n' {
in_line_comment = false;
}
previous = Some(ch);
i += 1;
continue;
}
if block_comment_depth > 0 {
if options.profile.supports_nested_block_comments && previous == Some('/') && ch == '*' {
block_comment_depth += 1;
// A nested opener's asterisk must not pair with a following slash as a close.
previous = None;
i += 1;
continue;
} else if previous == Some('*') && ch == '/' {
block_comment_depth -= 1;
}
previous = Some(ch);
i += 1;
continue;
}
if ch == '-' && next == Some('-') && dash_dash_starts_line_comment(options.profile, chars.get(i + 2).copied()) {
in_line_comment = true;
previous = Some(ch);
i += 1;
continue;
}
if options.profile.supports_hash_line_comments && ch == '#' {
in_line_comment = true;
previous = Some(ch);
i += 1;
continue;
}
if ch == '/' && next == Some('*') {
if is_mysql_executable_comment_start(&chars, i) {
return true;
}
block_comment_depth = 1;
// The opener's own asterisk must not read as a nested opener.
previous = None;
i += 1;
continue;
}
if !ch.is_whitespace() {
return true;
}
previous = Some(ch);
i += 1;
}
false
}
fn is_mysql_executable_comment_start(chars: &[char], start: usize) -> bool {
chars.get(start) == Some(&'/')
&& chars.get(start + 1) == Some(&'*')
&& (chars.get(start + 2) == Some(&'!')
|| (chars.get(start + 2) == Some(&'M') && chars.get(start + 3) == Some(&'!')))
}
#[cfg(test)]
fn split_sql_script(sql: &str) -> Result<Vec<String>, String> {
Ok(split_sql_statements(sql))
}
#[cfg(test)]
mod tests {
use crate::models::connection::DatabaseType;
use super::{
contains_or_fuzzy_match, dash_dash_starts_line_comment, decode_sql_file_bytes,
find_statement_at_cursor_for_database, fuzzy_filter_enabled, fuzzy_like_pattern_with_escape,
fuzzy_subsequence_match, optimize_sql_file_import_statements, prepare_sql_file_statement, split_sql_script,
split_sql_statement_ranges_with_options, split_sql_statements_for_database, starts_with_executable_sql_keyword,
starts_with_executable_sql_keyword_for_database, starts_with_oracle_style_routine_body, SqlDialectProfile,
SqlFileRequest, SqlFileStatementAction, SqlParsingOptions, SqlStatementSplitter,
};
#[test]
fn file_directive_boundaries_use_the_dialect_line_comment_rule() {
let mut mysql = SqlStatementSplitter::with_options(SqlParsingOptions::for_database_type(DatabaseType::Mysql));
mysql.push_chunk("--foo\n");
assert!(!mysql.is_at_file_directive_boundary());
let mut sqlite = SqlStatementSplitter::with_options(SqlParsingOptions::for_database_type(DatabaseType::Sqlite));
sqlite.push_chunk("--foo\n");
assert!(sqlite.is_at_file_directive_boundary());
let mut mysql_comment =
SqlStatementSplitter::with_options(SqlParsingOptions::for_database_type(DatabaseType::Mysql));
mysql_comment.push_chunk("-- real comment\n");
assert!(mysql_comment.is_at_file_directive_boundary());
}
#[test]
fn sql_file_request_keeps_failure_logging_on_when_the_field_is_absent() {
// Older frontends never send the flag; their imports must keep writing
// the failure log (dbx#11619), so the serde default is "enabled".
let base = serde_json::json!({
"executionId": "exec-1",
"connectionId": "conn-1",
"database": "shop",
"filePath": "/tmp/dump.sql",
"continueOnError": false
});
let legacy: SqlFileRequest = serde_json::from_value(base.clone()).unwrap();
assert!(legacy.failure_log_enabled);
let mut explicit_off = base;
explicit_off["failureLogEnabled"] = serde_json::Value::Bool(false);
let disabled: SqlFileRequest = serde_json::from_value(explicit_off).unwrap();
assert!(!disabled.failure_log_enabled);
}
#[test]
fn sql_file_request_schema_is_optional_and_round_trips_when_present() {
let legacy = serde_json::json!({
"executionId": "legacy",
"connectionId": "mysql-1",
"database": "app",
"filePath": "backup.sql",
"continueOnError": false
});
let legacy_request: SqlFileRequest = serde_json::from_value(legacy).unwrap();
assert_eq!(legacy_request.schema, None);
assert!(serde_json::to_value(&legacy_request).unwrap().get("schema").is_none());
let with_schema = serde_json::json!({
"executionId": "oracle-schema",
"connectionId": "oceanbase-oracle-1",
"database": "tenant_service",
"schema": "APP",
"filePath": "restore.sql",
"continueOnError": false
});
let request: SqlFileRequest = serde_json::from_value(with_schema).unwrap();
assert_eq!(request.database, "tenant_service");
assert_eq!(request.schema.as_deref(), Some("APP"));
assert_eq!(serde_json::to_value(request).unwrap()["schema"], "APP");
}
#[test]
fn fuzzy_subsequence_match_matches_ordered_characters() {
assert!(fuzzy_subsequence_match("system_user", "sysu"));
assert!(contains_or_fuzzy_match("user_order", "uo"));
assert!(!contains_or_fuzzy_match("alpha", "uo"));
}
#[test]
fn contains_or_fuzzy_match_skips_fuzzy_for_single_character_filters() {
assert!(fuzzy_filter_enabled("uo"));
assert!(!fuzzy_filter_enabled("u"));
assert!(contains_or_fuzzy_match("user_order", "u"));
assert!(!contains_or_fuzzy_match("orders", "u"));
}
#[test]
fn fuzzy_like_pattern_with_escape_keeps_wildcards_literal() {
let pattern = fuzzy_like_pattern_with_escape("user_%", |value| value.replace('%', "\\%").replace('_', "\\_"));
assert_eq!(pattern, "%u%s%e%r%\\_%\\%%");
}
#[test]
fn splits_semicolon_delimited_statements() {
assert_eq!(
split_sql_script("CREATE TABLE a(id int); INSERT INTO a VALUES (1);").unwrap(),
vec!["CREATE TABLE a(id int)", "INSERT INTO a VALUES (1)"]
);
}
#[test]
fn keeps_mybatis_comparison_entity_semicolons_inside_one_statement() {
// Issue #11699: the `;` of `>`/`<`/`'` closes an XML entity,
// not the statement.
let sql = "select * from course_user cu\nwhere cu.is_delete = 0\nand cu.create_time >= #{param.startDateTime}\nand cu.create_time <= #{param.endDateTime}\ngroup by cu.dept_id";
assert_eq!(split_sql_statements_for_database(sql, DatabaseType::Mysql), vec![sql.to_string()]);
assert_eq!(
split_sql_statements_for_database("SELECT a FROM t WHERE b >= 1;\nSELECT 2;", DatabaseType::Mysql),
vec!["SELECT a FROM t WHERE b >= 1".to_string(), "SELECT 2".to_string()]
);
// A bitwise AND before a real terminator must still split.
assert_eq!(
split_sql_statements_for_database("SELECT a FROM t WHERE a&b;", DatabaseType::Mysql),
vec!["SELECT a FROM t WHERE a&b".to_string()]
);
// Numeric character references and quoted entities stay intact too.
assert_eq!(
split_sql_statements_for_database("SELECT 'Tom & Jerry' AS n;\nSELECT 2;", DatabaseType::Mysql),
vec!["SELECT 'Tom & Jerry' AS n".to_string(), "SELECT 2".to_string()]
);
// Numeric character references: MySQL-family dialects read a bare `#` as a
// line comment, so exercise this on SQL Server (whose `#` is an identifier
// marker) where `'` reaches the splitter as code.
assert_eq!(
split_sql_statements_for_database("SELECT ' AS q;\nSELECT 2;", DatabaseType::SqlServer),
vec!["SELECT ' AS q".to_string(), "SELECT 2".to_string()]
);
}
#[test]
fn keeps_postgres_e_string_newlines_inside_batch_statement() {
let sql = "SELECT version();\n\nCREATE TABLE dbx_test (\n delimiter VARCHAR(32) DEFAULT E'\\n\\n'\n);";
assert_eq!(
split_sql_statements_for_database(sql, DatabaseType::Postgres),
vec![
"SELECT version()".to_string(),
"CREATE TABLE dbx_test (\n delimiter VARCHAR(32) DEFAULT E'\\n\\n'\n)".to_string(),
]
);
assert_eq!(
split_sql_statement_ranges_with_options(sql, SqlParsingOptions::for_database_type(DatabaseType::Postgres))
.into_iter()
.map(|range| range.text)
.collect::<Vec<_>>(),
vec![
"SELECT version()".to_string(),
"CREATE TABLE dbx_test (\n delimiter VARCHAR(32) DEFAULT E'\\n\\n'\n)".to_string(),
]
);
}
#[test]
fn delimiter_line_after_unterminated_statement_merges() {
// A DELIMITER line is only honored as a client directive when no executable
// SQL is pending, so an unterminated statement swallows the line instead of
// splitting off and switching the delimiter.
let sql = "SELECT 1\nDELIMITER $$";
assert_eq!(
split_sql_statements_for_database(sql, DatabaseType::Postgres),
vec!["SELECT 1\nDELIMITER $$".to_string()]
);
assert_eq!(
split_sql_statement_ranges_with_options(sql, SqlParsingOptions::for_database_type(DatabaseType::Postgres))
.into_iter()
.map(|range| range.text)
.collect::<Vec<_>>(),
vec!["SELECT 1\nDELIMITER $$".to_string()]
);
let sql = "SELECT 1\nDELIMITER $$\nSELECT 2$$";
assert_eq!(
split_sql_statements_for_database(sql, DatabaseType::Postgres),
vec!["SELECT 1\nDELIMITER $$\nSELECT 2$$".to_string()]
);
assert_eq!(
split_sql_statement_ranges_with_options(sql, SqlParsingOptions::for_database_type(DatabaseType::Postgres))
.into_iter()
.map(|range| range.text)
.collect::<Vec<_>>(),
vec!["SELECT 1\nDELIMITER $$\nSELECT 2$$".to_string()]
);
}
#[test]
fn mysql_split_skips_comment_only_statement() {
assert!(split_sql_statements_for_database("-- DBX SQL preview crash reproducer\n\n;", DatabaseType::Mysql)
.is_empty());
}
#[test]
fn mysql_dash_dash_without_trailing_space_is_not_a_comment_per_issue_5382() {
// MySQL requires `--` to be followed by whitespace to start a line comment;
// `5--1` is subtraction (5 - -1), not a comment.
assert_eq!(split_sql_statements_for_database("SELECT 5--1;", DatabaseType::Mysql), vec!["SELECT 5--1"]);
}
#[test]
fn mysql_dash_dash_without_trailing_space_does_not_merge_following_statements_per_issue_5382() {
let statements = split_sql_statements_for_database(
"INSERT INTO t VALUES (1, 5--1);\nINSERT INTO t VALUES (2, 10);",
DatabaseType::Mysql,
);
assert_eq!(statements, vec!["INSERT INTO t VALUES (1, 5--1)", "INSERT INTO t VALUES (2, 10)"]);
}
#[test]
fn mysql_dash_dash_with_trailing_space_still_starts_a_line_comment() {
// The comment text itself is preserved verbatim in the following statement's
// buffer (comments are never stripped, just protected from delimiter parsing) —
// this only asserts the embedded `;` inside the comment doesn't split early.
let statements = split_sql_statements_for_database(
"INSERT INTO t VALUES (1); -- trailing comment; with semicolon\nINSERT INTO t VALUES (2);",
DatabaseType::Mysql,
);
assert_eq!(
statements,
vec!["INSERT INTO t VALUES (1)", "-- trailing comment; with semicolon\nINSERT INTO t VALUES (2)"]
);
}
#[test]
fn dash_dash_requires_space_only_for_mysql_compatible_profiles() {
let mysql_profile = SqlDialectProfile::for_database_type(DatabaseType::Mysql);
assert!(dash_dash_starts_line_comment(mysql_profile, Some(' ')));
assert!(dash_dash_starts_line_comment(mysql_profile, Some('\n')));
assert!(dash_dash_starts_line_comment(mysql_profile, Some('\u{1}')));
assert!(dash_dash_starts_line_comment(mysql_profile, None));
assert!(!dash_dash_starts_line_comment(mysql_profile, Some('1')));
// Postgres, Oracle/DM, SQL Server, ... treat `--` as a comment opener unconditionally.
let default_profile = SqlDialectProfile::default();
assert!(dash_dash_starts_line_comment(default_profile, Some('1')));
assert!(dash_dash_starts_line_comment(default_profile, None));
}
#[test]
fn mysql_dash_dash_opener_without_space_can_span_chunks() {
let mut splitter = SqlStatementSplitter::with_options(SqlParsingOptions::mysql_compatible());
assert_eq!(splitter.push_chunk("SELECT 5-"), Vec::<String>::new());
assert_eq!(splitter.push_chunk("-1;"), vec!["SELECT 5--1"]);
assert_eq!(splitter.finish(), Vec::<String>::new());
}
#[test]
fn mysql_complete_dash_dash_sequence_waits_for_the_next_chunk() {
let mut splitter = SqlStatementSplitter::with_options(SqlParsingOptions::mysql_compatible());
assert_eq!(splitter.push_chunk("SELECT 5--"), Vec::<String>::new());
assert_eq!(splitter.push_chunk("1;\nSELECT 2;"), vec!["SELECT 5--1", "SELECT 2"]);
assert_eq!(splitter.finish(), Vec::<String>::new());
}
#[test]
fn mysql_control_character_after_dashes_starts_a_line_comment() {
let statements = split_sql_statements_for_database("SELECT 1--\u{1}; hidden\nSELECT 2;", DatabaseType::Mysql);
assert_eq!(statements, vec!["SELECT 1--\u{1}; hidden\nSELECT 2"]);
}
#[test]
fn mysql_keyword_detection_skips_comment_only_input() {
assert!(!starts_with_executable_sql_keyword_for_database(
"-- DBX SQL preview crash reproducer\n\n",
&["SELECT"],
DatabaseType::Mysql,
));
}
#[test]
fn decodes_utf8_bom_sql_file_bytes_without_bom_statement_prefix() {
let sql = decode_sql_file_bytes(b"\xEF\xBB\xBFCREATE TABLE t(id int);").unwrap();
assert_eq!(sql, "CREATE TABLE t(id int);");
}
#[test]
fn decodes_gbk_sql_file_bytes_before_execution() {
let bytes = b"INSERT INTO t VALUES ('\xD6\xD0\xCE\xC4');";
let sql = decode_sql_file_bytes(bytes).unwrap();
assert_eq!(sql, "INSERT INTO t VALUES ('中文');");
}
#[test]
fn decodes_utf16le_bom_sql_file_bytes() {
let bytes = [
0xFF, 0xFE, b'S', 0x00, b'E', 0x00, b'L', 0x00, b'E', 0x00, b'C', 0x00, b'T', 0x00, b' ', 0x00, b'1', 0x00,
b';', 0x00,
];
let sql = decode_sql_file_bytes(&bytes).unwrap();
assert_eq!(sql, "SELECT 1;");
}
#[test]
fn ignores_semicolons_inside_quotes_and_comments() {
let sql = "\
INSERT INTO logs VALUES ('a;b', \"c;d\", `weird;name`);\n\
-- comment ; ignored\n\
/* block ; ignored */\n\
SELECT 1;";
assert_eq!(
split_sql_script(sql).unwrap(),
vec![
"INSERT INTO logs VALUES ('a;b', \"c;d\", `weird;name`)",
"-- comment ; ignored\n/* block ; ignored */\nSELECT 1",
]
);
}
#[test]
fn closes_mysql_string_after_even_trailing_backslashes() {
let sql = r#"CREATE TABLE paths (value varchar(100) COMMENT 'Windows path\\'); DROP TABLE paths;"#;
assert_eq!(
split_sql_statements_for_database(sql, DatabaseType::Mysql),
vec![r#"CREATE TABLE paths (value varchar(100) COMMENT 'Windows path\\')"#, "DROP TABLE paths"]
);
}
#[test]
fn cursor_statement_closes_mysql_string_after_escaped_backslash() {
let sql = "SELECT 'a\\\\';\nSELECT 2;";
let cursor = sql.find("SELECT 2").unwrap() + 3;
assert_eq!(find_statement_at_cursor_for_database(sql, cursor, DatabaseType::Mysql), "SELECT 2");
assert_eq!(find_statement_at_cursor_for_database(sql, 3, DatabaseType::Mysql), "SELECT 'a\\\\'");
}
#[test]
fn cursor_statement_treats_mysql_dash_dash_without_space_as_minus() {
let sql = "SELECT 5--1;\nSELECT 2;";
let cursor = sql.find("SELECT 2").unwrap() + 3;
assert_eq!(find_statement_at_cursor_for_database(sql, cursor, DatabaseType::Mysql), "SELECT 2");
assert_eq!(find_statement_at_cursor_for_database(sql, 3, DatabaseType::Mysql), "SELECT 5--1");
}
#[test]
fn keeps_mysql_string_open_after_odd_trailing_backslashes() {
let sql = r#"INSERT INTO notes VALUES ('it\'s; still one value'); SELECT 1;"#;
assert_eq!(
split_sql_statements_for_database(sql, DatabaseType::Mysql),
vec![r#"INSERT INTO notes VALUES ('it\'s; still one value')"#, "SELECT 1"]
);
}
#[test]
fn closes_mysql_string_after_even_trailing_backslashes_across_chunks() {
let mut splitter =
SqlStatementSplitter::with_options(SqlParsingOptions::for_database_type(DatabaseType::Mysql));
assert!(splitter.push_chunk("CREATE TABLE paths (value varchar(100) COMMENT 'Windows path\\").is_empty());
assert_eq!(
splitter.push_chunk("\\'); DROP TABLE paths;"),
vec![r#"CREATE TABLE paths (value varchar(100) COMMENT 'Windows path\\')"#, "DROP TABLE paths"]
);
assert!(splitter.finish().is_empty());
}
#[test]
fn postgres_plain_string_ends_before_a_trailing_backslash() {
let sql = r#"SELECT 'dir\'; DELETE FROM users;"#;
assert_eq!(
split_sql_statements_for_database(sql, DatabaseType::Postgres),
vec![r#"SELECT 'dir\'"#, "DELETE FROM users"]
);
}
#[test]
fn vastbase_plain_string_ends_before_a_trailing_backslash() {
let sql = r#"SELECT 'dir\'; DELETE FROM users;"#;
assert_eq!(
split_sql_statements_for_database(sql, DatabaseType::Vastbase),
vec![r#"SELECT 'dir\'"#, "DELETE FROM users"]
);
}
#[test]
fn vastbase_escape_string_keeps_a_backslash_escaped_quote() {
let sql = r#"COMMENT ON COLUMN "s"."t"."c" IS E'C:\\tmp\n\t\'';"#;
assert_eq!(
split_sql_statements_for_database(sql, DatabaseType::Vastbase),
vec![r#"COMMENT ON COLUMN "s"."t"."c" IS E'C:\\tmp\n\t\''"#]
);
}
#[test]
fn duckdb_escape_strings_match_the_existing_expectations() {
let sql = r#"SELECT E'it\'s;ok'; SELECT e'path\\;name'; SELECT 2"#;
assert_eq!(
split_sql_statements_for_database(sql, DatabaseType::DuckDb),
vec![r#"SELECT E'it\'s;ok'"#, r#"SELECT e'path\\;name'"#, "SELECT 2"]
);
}
#[test]
fn duckdb_plain_strings_keep_windows_paths_and_following_statements_separate() {
let sql = r#"SET VARIABLE path = 'C:\path\'; SET VARIABLE next = 'value'; SELECT 1;"#;
assert_eq!(
split_sql_statements_for_database(sql, DatabaseType::DuckDb),
vec![r#"SET VARIABLE path = 'C:\path\'"#, "SET VARIABLE next = 'value'", "SELECT 1"]
);
}
#[test]
fn postgres_escape_string_keeps_a_backslash_escaped_quote() {
let sql = r#"SELECT E'it\'s; still one value'; SELECT 1;"#;
assert_eq!(
split_sql_statements_for_database(sql, DatabaseType::Postgres),
vec![r#"SELECT E'it\'s; still one value'"#, "SELECT 1"]
);
}
#[test]
fn cursor_statement_keeps_postgres_delete_after_a_trailing_backslash() {
let sql = "SELECT 'dir\\';\nDELETE FROM users;";
let cursor = sql.find("DELETE FROM users").unwrap() + 3;
assert_eq!(find_statement_at_cursor_for_database(sql, cursor, DatabaseType::Postgres), "DELETE FROM users");
}
#[test]
fn cursor_statement_keeps_postgres_escape_string_quote_together() {
let sql = "SELECT E'it\\'s; x';\nSELECT 2;";
let cursor = sql.find("SELECT 2").unwrap() + 3;
assert_eq!(find_statement_at_cursor_for_database(sql, cursor, DatabaseType::Postgres), "SELECT 2");
assert_eq!(find_statement_at_cursor_for_database(sql, 3, DatabaseType::Postgres), r#"SELECT E'it\'s; x'"#);
}
#[test]
fn postgres_nested_block_comments_keep_commented_out_statements_out() {
// The inner comment closes first, so without nesting the UPDATE behind the
// semicolon becomes a statement of its own and reaches the executor.
let sql = "/* a\n/* b */ ;\nUPDATE prod.orders SET paid = 0;\n*/\nSELECT 1;";
let statements = split_sql_statements_for_database(sql, DatabaseType::Postgres);
assert!(!statements.iter().any(|statement| statement.starts_with("UPDATE")), "{statements:?}");
assert!(statements.iter().any(|statement| statement.contains("SELECT 1")), "{statements:?}");
}
#[test]
fn sqlserver_nested_block_comments_keep_commented_out_statements_out() {
let sql = "/* a\n/* b */ ;\nUPDATE prod.orders SET paid = 0;\n*/\nSELECT 1;";
let statements = split_sql_statements_for_database(sql, DatabaseType::SqlServer);
assert!(!statements.iter().any(|statement| statement.starts_with("UPDATE")), "{statements:?}");
assert!(statements.iter().any(|statement| statement.contains("SELECT 1")), "{statements:?}");
}
#[test]
fn nested_opener_slash_asterisk_slash_does_not_close_the_comment() {
// The lexer consumes `/*` atomically, so the `/` right after a nested opener
// is literal text and must not pair with the preceding `*` to close it.
let sql = "/* /*/ */ SELECT 1;";
let statements = split_sql_statements_for_database(sql, DatabaseType::Postgres);
assert!(!statements.iter().any(|statement| statement.trim_start().starts_with("SELECT")), "{statements:?}");
}
#[test]
fn clickhouse_block_comments_nest() {
let sql = "/* a\n/* b */ ;\nUPDATE prod.orders SET paid = 0;\n*/\nSELECT 1;";
let statements = split_sql_statements_for_database(sql, DatabaseType::ClickHouse);
assert!(!statements.iter().any(|statement| statement.starts_with("UPDATE")), "{statements:?}");
assert!(statements.iter().any(|statement| statement.contains("SELECT 1")), "{statements:?}");
}
#[test]
fn mysql_block_comments_do_not_nest() {
let sql = "/* a /* b */ SELECT 1; SELECT 2;";
// MySQL documents that block comments do not nest, so the inner `*/` ends
// the comment and the statements behind it stay separate.
assert_eq!(
split_sql_statements_for_database(sql, DatabaseType::Mysql),
vec!["/* a /* b */ SELECT 1", "SELECT 2"]
);
}
#[test]
fn cursor_statement_ignores_statements_inside_a_nested_comment() {
let sql = "/* a\n/* b */ ;\nUPDATE prod.orders SET paid = 0;\n*/\nSELECT 1;";
let cursor = sql.find("UPDATE").unwrap() + 3;
let statement = find_statement_at_cursor_for_database(sql, cursor, DatabaseType::Postgres);
assert!(!statement.contains("UPDATE prod.orders"), "{statement}");
}
#[test]
fn oracle_alternative_quoting_keeps_apostrophes_and_semicolons_inside_the_literal() {
let sql = "SELECT q'[it's; still one value]' FROM dual; SELECT 1 FROM dual;";
assert_eq!(
split_sql_statements_for_database(sql, DatabaseType::Oracle),
vec!["SELECT q'[it's; still one value]' FROM dual", "SELECT 1 FROM dual"]
);
}
#[test]
fn oracle_national_alternative_quoting_preserves_statement_boundaries() {
for database in [DatabaseType::Oracle, DatabaseType::OceanbaseOracle] {
let sql = "UPDATE events SET name = nq'{it's; DELETE FROM events; --}'; SELECT 1 FROM dual;";
assert_eq!(
split_sql_statements_for_database(sql, database),
vec!["UPDATE events SET name = nq'{it's; DELETE FROM events; --}'", "SELECT 1 FROM dual"]
);
let sql = "UPDATE events SET name = nq'[a'b]'; DELETE FROM events; --'";
assert_eq!(
split_sql_statements_for_database(sql, database),
vec!["UPDATE events SET name = nq'[a'b]'", "DELETE FROM events"]
);
}
}
#[test]
fn oracle_alternative_quoting_supports_every_delimiter_form() {
for oracle_literal in [
"q'{a;b}'",
"q'(a;b)'",
"q'<a;b>'",
"q'!a;b!'",
"q'Aa;bA'",
"Q'[i's; x]'",
"nq'[i's; x]'",
"NQ'<i's; x>'",
"nq'!i's; x!'",
"nQ'(i's; x)'",
] {
let sql = format!("SELECT {oracle_literal} FROM dual; SELECT 1 FROM dual;");
assert_eq!(
split_sql_statements_for_database(&sql, DatabaseType::Oracle),
vec![format!("SELECT {oracle_literal} FROM dual"), "SELECT 1 FROM dual".to_string()],
"{oracle_literal}"
);
}
}
#[test]
fn oracle_alternative_quoting_stays_intact_across_chunk_boundaries() {
for literal in ["q'[it's; one]'", "nq'{it's; one}'", "NQ'<it's; one>'", "nQ'!it's; one!'"] {
let statement = format!("UPDATE events SET name = {literal}");
let sql = format!("{statement}; DELETE FROM events;");
for boundary in 0..=sql.len() {
let mut splitter =
SqlStatementSplitter::with_options(SqlParsingOptions::for_database_type(DatabaseType::Oracle));
let mut statements = splitter.push_chunk(&sql[..boundary]);
statements.extend(splitter.push_chunk(&sql[boundary..]));
statements.extend(splitter.finish());
assert_eq!(
statements,
vec![statement.clone(), "DELETE FROM events".to_string()],
"boundary {boundary}"
);
}
}
}
#[test]
fn oracle_alternative_quoting_requires_an_introducer_token() {
for prefix in ["aq", "anq", "_nq", "column$nq"] {
let sql = format!("SELECT {prefix}'[a'b]'; DELETE FROM events; --'");
assert_eq!(split_sql_statements_for_database(&sql, DatabaseType::Oracle), vec![sql]);
}
}
#[test]
fn non_oracle_dialects_do_not_interpret_national_alternative_quoting() {
let sql = "SELECT nq'[a'b]'; DELETE FROM events; --'";
for database in [DatabaseType::Mysql, DatabaseType::Postgres, DatabaseType::Sqlite, DatabaseType::SqlServer] {
assert_eq!(split_sql_statements_for_database(sql, database), vec![sql]);
}
}
#[test]
fn oceanbase_oracle_unsupported_quoting_does_not_hide_statement_boundaries() {
for (sql, expected) in [
("UPDATE t SET x=$$;DELETE FROM t;$$", vec!["UPDATE t SET x=$$", "DELETE FROM t", "$$"]),
("UPDATE t SET x='a\\'; DELETE FROM t; --'", vec!["UPDATE t SET x='a\\'", "DELETE FROM t"]),
("UPDATE t SET x=`a;DELETE FROM t;`", vec!["UPDATE t SET x=`a", "DELETE FROM t", "`"]),
(
"DELIMITER $$\nUPDATE t SET x=1; DELETE FROM t;$$",
vec!["DELIMITER $$\nUPDATE t SET x=1", "DELETE FROM t", "$$"],
),
] {
assert_eq!(split_sql_statements_for_database(sql, DatabaseType::OceanbaseOracle), expected, "{sql}");
}
}
#[test]
fn oceanbase_oracle_cursor_uses_its_own_quote_boundaries() {
for sql in [
"UPDATE t SET x=$$; DELETE FROM t;$$",
"UPDATE t SET x='a\\'; DELETE FROM t; --'",
"UPDATE t SET x=`a; DELETE FROM t;`",
"DELIMITER $$\nUPDATE t SET x=1; DELETE FROM t;$$",
] {
let cursor = sql.find("DELETE").unwrap() + 3;
assert_eq!(
find_statement_at_cursor_for_database(sql, cursor, DatabaseType::OceanbaseOracle),
"DELETE FROM t",
"{sql}"
);
}
for sql in [
"SELECT 'a\\'\n\n\nUPDATE t SET x=2;",
"SELECT `x\n\n\nUPDATE t SET x=2;",
"SELECT $$\n\n\nUPDATE t SET x=2;",
] {
let cursor = sql.find("UPDATE").unwrap() + 3;
assert_eq!(
find_statement_at_cursor_for_database(sql, cursor, DatabaseType::OceanbaseOracle),
"UPDATE t SET x=2",
"{sql}"
);
}
}
#[test]
fn oceanbase_oracle_quote_rules_do_not_change_other_dialects() {
for database in [DatabaseType::Oracle, DatabaseType::Dameng, DatabaseType::Gaussdb, DatabaseType::Mysql] {
for sql in ["UPDATE t SET x=$$;DELETE FROM t;$$", "UPDATE t SET x=`a;DELETE FROM t;`"] {
assert_eq!(split_sql_statements_for_database(sql, database), vec![sql]);
}
}
let statement = "SELECT E'a\\'\n\n\nDELETE FROM t' FROM t";
let sql = format!("{statement}; SELECT 2;");
for database in [DatabaseType::Postgres, DatabaseType::Gaussdb] {
let cursor = sql.find("DELETE").unwrap() + 3;
assert_eq!(find_statement_at_cursor_for_database(&sql, cursor, database), statement);
}
}
#[test]
fn dameng_alternative_quoting_matches_oracle() {
let sql = "SELECT q'[it's; x]' FROM dual; SELECT 1 FROM dual;";
assert_eq!(
split_sql_statements_for_database(sql, DatabaseType::Dameng),
vec!["SELECT q'[it's; x]' FROM dual", "SELECT 1 FROM dual"]
);
}
#[test]
fn mysql_treats_q_quotes_as_plain_strings() {
let sql = "SELECT q'[a;b]' FROM t; SELECT 1;";
assert_eq!(
split_sql_statements_for_database(sql, DatabaseType::Mysql),
vec!["SELECT q'[a;b]' FROM t", "SELECT 1"]
);
}
#[test]
fn cursor_statement_keeps_oracle_alternative_quoting_together() {
let sql = "SELECT q'[it's; still one value]' FROM dual;\nSELECT 1 FROM dual;";
let cursor = sql.find("SELECT 1").unwrap() + 3;
assert_eq!(find_statement_at_cursor_for_database(sql, cursor, DatabaseType::Oracle), "SELECT 1 FROM dual");
}
#[test]
fn oracle_like_cursor_recognizes_blank_line_boundary_after_backslash() {
let sql = "SELECT 'a\\'\n\n\nUPDATE t SET x=2;";
for database in [
DatabaseType::Oracle,
DatabaseType::OceanbaseOracle,
DatabaseType::Dameng,
DatabaseType::Yashandb,
DatabaseType::Oscar,
] {
let cursor = sql.find("UPDATE").unwrap() + 3;
assert_eq!(
find_statement_at_cursor_for_database(sql, cursor, database),
"UPDATE t SET x=2",
"{database:?}"
);
}
}
#[test]
fn cursor_statement_keeps_national_alternative_quoting_and_blank_lines_together() {
let statement = "UPDATE events SET name = NQ'<it's;\n\n\nDELETE FROM events; -->' WHERE id = 1";
let sql = format!("{statement};\nSELECT 1 FROM dual;");
for database in [DatabaseType::Oracle, DatabaseType::OceanbaseOracle] {
let cursor = sql.find("DELETE").unwrap() + 3;
assert_eq!(find_statement_at_cursor_for_database(&sql, cursor, database), statement);
let cursor = sql.find("SELECT 1").unwrap() + 3;
assert_eq!(find_statement_at_cursor_for_database(&sql, cursor, database), "SELECT 1 FROM dual");
}
}
#[test]
fn emits_trailing_statement_without_semicolon() {
assert_eq!(
split_sql_script("CREATE TABLE a(id int);\nINSERT INTO a VALUES (1)").unwrap(),
vec!["CREATE TABLE a(id int)", "INSERT INTO a VALUES (1)"]
);
}
#[test]
fn line_comment_openers_can_span_chunks() {
let mut splitter = SqlStatementSplitter::default();
assert_eq!(splitter.push_chunk("SELECT 1; -"), vec!["SELECT 1"]);
assert_eq!(splitter.push_chunk("- comment ; ignored\nSELECT 2;"), vec!["-- comment ; ignored\nSELECT 2"]);
assert_eq!(splitter.finish(), Vec::<String>::new());
}
#[test]
fn block_comment_openers_can_span_chunks() {
let mut splitter = SqlStatementSplitter::default();
assert_eq!(splitter.push_chunk("SELECT 1; /"), vec!["SELECT 1"]);
assert_eq!(splitter.push_chunk("* comment ; ignored */\nSELECT 2;"), vec!["/* comment ; ignored */\nSELECT 2"]);
assert_eq!(splitter.finish(), Vec::<String>::new());
}
#[test]
fn skips_comment_only_tail_after_statement() {
assert_eq!(
split_sql_script("CREATE TABLE a(id int); -- done\n/* no more sql */").unwrap(),
vec!["CREATE TABLE a(id int)"]
);
}
#[test]
fn skips_comment_only_statement_with_semicolon() {
assert_eq!(
split_sql_script("-- insert sqluser.tb_a values (6,'006','测试6','无');").unwrap(),
Vec::<String>::new()
);
assert!(!super::has_executable_sql("-- insert sqluser.tb_a values (6,'006','测试6','无');"));
}
#[test]
fn keeps_postgres_dollar_quoted_function_body_together() {
let sql = "\
CREATE FUNCTION bump_counter()\n\
RETURNS trigger AS $$\n\
BEGIN\n\
PERFORM 1;\n\
RETURN NEW;\n\
END;\n\
$$ LANGUAGE plpgsql;\n\
SELECT 1;";
assert_eq!(
split_sql_script(sql).unwrap(),
vec![
"CREATE FUNCTION bump_counter()\nRETURNS trigger AS $$\nBEGIN\nPERFORM 1;\nRETURN NEW;\nEND;\n$$ LANGUAGE plpgsql",
"SELECT 1",
]
);
}
#[test]
fn keeps_mysql_executable_comments_as_statements() {
assert_eq!(
split_sql_script("/*!40101 SET @OLD_CHARACTER_SET_CLIENT=@@CHARACTER_SET_CLIENT */;\nSELECT 1;",).unwrap(),
vec!["/*!40101 SET @OLD_CHARACTER_SET_CLIENT=@@CHARACTER_SET_CLIENT */", "SELECT 1",]
);
}
#[test]
fn detects_result_set_keyword_after_comments() {
assert!(starts_with_executable_sql_keyword("-- comment\nselect * from users;", &["SELECT"]));
assert!(starts_with_executable_sql_keyword(
"/* comment */\nWITH rows AS (SELECT 1) SELECT * FROM rows;",
&["WITH"]
));
assert!(!starts_with_executable_sql_keyword("-- comment only\n", &["SELECT"]));
}
#[test]
fn detects_mysql_executable_comment_keyword() {
assert!(starts_with_executable_sql_keyword("/*!40101 SELECT 1 */", &["SELECT"]));
assert!(starts_with_executable_sql_keyword("/*M! SELECT 1 */", &["SELECT"]));
}
#[test]
fn describe_keyword_detection_accepts_desc_shorthand() {
assert!(starts_with_executable_sql_keyword("DESC users", &["DESCRIBE"]));
assert!(starts_with_executable_sql_keyword("-- comment\nDESC users", &["DESCRIBE"]));
}
#[test]
fn detects_keyword_after_parentheses() {
assert!(starts_with_executable_sql_keyword("(SELECT 1)", &["SELECT"]));
assert!(starts_with_executable_sql_keyword(" ( SELECT 1 )", &["SELECT"]));
assert!(starts_with_executable_sql_keyword("((SELECT 1))", &["SELECT"]));
assert!(starts_with_executable_sql_keyword("(SELECT * FROM users)", &["SELECT"]));
assert!(starts_with_executable_sql_keyword("(INSERT INTO users VALUES (1))", &["INSERT"]));
assert!(starts_with_executable_sql_keyword("/* comment */(UPDATE users SET name = 'test')", &["UPDATE"]));
}
#[test]
fn mysql_hash_comments_are_ignored_for_keyword_detection() {
assert!(starts_with_executable_sql_keyword_for_database(
"# comment only for mysql\nSELECT 1",
&["SELECT"],
DatabaseType::Mysql
));
assert!(!starts_with_executable_sql_keyword("# comment only for mysql\nSELECT 1", &["SELECT"]));
}
#[test]
fn prepares_mysql_executable_comments_for_mysql_compatible_imports() {
assert_eq!(
prepare_sql_file_statement("/*!40101 SET NAMES utf8mb4 */", &DatabaseType::Mysql, None),
SqlFileStatementAction::Execute("SET NAMES utf8mb4".to_string())
);
}
#[test]
fn skips_mysql_key_toggle_comments_for_mysql_compatible_imports() {
assert_eq!(
prepare_sql_file_statement(" /*!40000 ALTER TABLE `dd_admin` ENABLE KEYS */", &DatabaseType::Mysql, None),
SqlFileStatementAction::Skip
);
assert_eq!(
prepare_sql_file_statement("/*!40000 ALTER TABLE `dd_admin` DISABLE KEYS */", &DatabaseType::Mysql, None),
SqlFileStatementAction::Skip
);
}
#[test]
fn skips_mysql_lock_table_statements_for_mysql_compatible_imports() {
assert_eq!(
prepare_sql_file_statement("LOCK TABLES `dd_geo_json` WRITE", &DatabaseType::Mysql, None),
SqlFileStatementAction::Skip
);
assert_eq!(
prepare_sql_file_statement("UNLOCK TABLES", &DatabaseType::Mysql, None),
SqlFileStatementAction::Skip
);
assert_eq!(
prepare_sql_file_statement(
"-- Dumping data for table `dd_geo_json`\nLOCK TABLES `dd_geo_json` WRITE",
&DatabaseType::Mysql,
None
),
SqlFileStatementAction::Skip
);
}
#[test]
fn skips_mysql_session_restore_statements_for_mysql_compatible_imports() {
assert_eq!(
prepare_sql_file_statement(
"/*!40101 SET character_set_client = @saved_cs_client */",
&DatabaseType::Mysql,
None
),
SqlFileStatementAction::Skip
);
assert_eq!(
prepare_sql_file_statement("/*!40103 SET TIME_ZONE=@OLD_TIME_ZONE */", &DatabaseType::Mysql, None),
SqlFileStatementAction::Skip
);
assert_eq!(
prepare_sql_file_statement("SET FOREIGN_KEY_CHECKS=@OLD_FOREIGN_KEY_CHECKS", &DatabaseType::Mysql, None),
SqlFileStatementAction::Skip
);
assert_eq!(
prepare_sql_file_statement(
"/*!40101 SET @saved_cs_client = @@character_set_client */",
&DatabaseType::Mysql,
None
),
SqlFileStatementAction::Execute("SET @saved_cs_client = @@character_set_client".to_string())
);
}
#[test]
fn skips_mysql_executable_comments_for_non_mysql_imports() {
assert_eq!(
prepare_sql_file_statement(
"/*!40101 SET character_set_client = @saved_cs_client */",
&DatabaseType::Postgres,
None
),
SqlFileStatementAction::Skip
);
}
#[test]
fn prepares_postgres_pg_dump_restrict_guard_with_following_sql() {
assert_eq!(
prepare_sql_file_statement(
"-- PostgreSQL database dump\n\n\\restrict PreICrqH3VX69RwvK06tauVFPoZ3tThfV0y2zIFL2uSfhfNPSu1pWtu1QTudUAc\n\nSET statement_timeout = 0",
&DatabaseType::Postgres,
None
),
SqlFileStatementAction::Execute("-- PostgreSQL database dump\n\n\nSET statement_timeout = 0".to_string())
);
}
#[test]
fn skips_standalone_postgres_pg_dump_unrestrict_guard() {
assert_eq!(
prepare_sql_file_statement(
"--\n-- PostgreSQL database dump complete\n--\n\n\\unrestrict PreICrqH3VX69RwvK06tauVFPoZ3tThfV0y2zIFL2uSfhfNPSu1pWtu1QTudUAc",
&DatabaseType::Postgres,
None
),
SqlFileStatementAction::Skip
);
}
#[test]
fn keeps_other_postgres_psql_commands_unsupported() {
assert_eq!(
prepare_sql_file_statement("\\copy users FROM 'users.csv'", &DatabaseType::Postgres, None),
SqlFileStatementAction::Execute("\\copy users FROM 'users.csv'".to_string())
);
assert_eq!(
prepare_sql_file_statement("\\restrict invalid-key", &DatabaseType::Postgres, None),
SqlFileStatementAction::Execute("\\restrict invalid-key".to_string())
);
}
#[test]
fn optimizes_adjacent_postgres_single_row_inserts_into_multi_row_insert() {
let statements = vec![
"INSERT INTO users (id, name) VALUES (1, 'Ada')".to_string(),
"insert into users (id, name) values (2, 'Linus')".to_string(),
"SELECT 1".to_string(),
];
let optimized = optimize_sql_file_import_statements(&statements, Some(DatabaseType::Postgres), None);
assert_eq!(optimized.len(), 2);
assert_eq!(optimized[0].source_statement_count, 2);
assert_eq!(optimized[0].sql, "INSERT INTO users (id, name) VALUES\n(1, 'Ada'),\n(2, 'Linus')");
assert_eq!(optimized[1].sql, "SELECT 1");
}
#[test]
fn optimizes_adjacent_oracle_family_inserts_into_insert_all() {
let statements = vec![
"Insert into SYS_QUESTION (ID, NAME) values ('001', '市场监管')".to_string(),
"Insert into SYS_QUESTION (ID, NAME) values ('002', '网络交易')".to_string(),
];
for db_type in [DatabaseType::Oracle, DatabaseType::OceanbaseOracle] {
let optimized = optimize_sql_file_import_statements(&statements, Some(db_type), None);
assert_eq!(optimized.len(), 1, "{db_type:?}");
assert_eq!(optimized[0].source_statement_count, 2, "{db_type:?}");
assert_eq!(
optimized[0].sql,
"INSERT ALL\n\
INTO SYS_QUESTION (ID, NAME) VALUES ('001', '市场监管')\n\
INTO SYS_QUESTION (ID, NAME) VALUES ('002', '网络交易')\n\
SELECT 1 FROM dual",
"{db_type:?}"
);
}
}
#[test]
fn keeps_single_oracle_insert_as_written() {
let statements = vec!["INSERT INTO users (id) VALUES (1)".to_string()];
let optimized = optimize_sql_file_import_statements(&statements, Some(DatabaseType::OceanbaseOracle), None);
assert_eq!(optimized.len(), 1);
assert_eq!(optimized[0].sql, statements[0]);
}
#[test]
fn keeps_mysql_insert_statements_separate_to_preserve_session_semantics() {
let statements = vec![
"INSERT INTO parents (name) VALUES ('first')".to_string(),
"INSERT INTO parents (name) VALUES ('second')".to_string(),
"INSERT INTO children (parent_id) VALUES (LAST_INSERT_ID())".to_string(),
];
for (db_type, driver_profile) in
[(DatabaseType::Mysql, None), (DatabaseType::Jdbc, Some("mariadb")), (DatabaseType::Doris, None)]
{
let optimized = optimize_sql_file_import_statements(&statements, Some(db_type), driver_profile);
assert_eq!(optimized.len(), 3, "{db_type:?}/{driver_profile:?}");
assert!(
optimized.iter().all(|statement| statement.source_statement_count == 1),
"{db_type:?}/{driver_profile:?}"
);
assert_eq!(
optimized.iter().map(|statement| statement.sql.as_str()).collect::<Vec<_>>(),
statements,
"{db_type:?}/{driver_profile:?}"
);
}
}
#[test]
fn keeps_existing_mysql_multi_row_insert_unchanged() {
let statements = vec!["INSERT INTO users (id) VALUES (1), (2)".to_string()];
let optimized = optimize_sql_file_import_statements(&statements, Some(DatabaseType::Mysql), None);
assert_eq!(optimized.len(), 1);
assert_eq!(optimized[0].source_statement_count, 1);
assert_eq!(optimized[0].sql, statements[0]);
}
#[test]
fn keeps_insert_batches_separate_for_different_targets_or_suffixes() {
let statements = vec![
"INSERT INTO users (id) VALUES (1)".to_string(),
"INSERT INTO teams (id) VALUES (1)".to_string(),
"INSERT INTO users (id) VALUES (2) RETURNING id".to_string(),
];
let optimized = optimize_sql_file_import_statements(&statements, Some(DatabaseType::Postgres), None);
assert_eq!(optimized.len(), 3);
assert!(optimized.iter().all(|statement| statement.source_statement_count == 1));
}
#[test]
fn mysql_dump_import_keeps_skips_and_insert_statement_boundaries() {
let statements = vec![
"LOCK TABLES `users` WRITE".to_string(),
"INSERT INTO `users` VALUES (1)".to_string(),
"INSERT INTO `users` VALUES (2)".to_string(),
"UNLOCK TABLES".to_string(),
];
let optimized = optimize_sql_file_import_statements(&statements, Some(DatabaseType::Mysql), None);
assert_eq!(optimized.len(), 4);
assert_eq!(optimized[0].kind, super::SqlFileImportStatementKind::Skip);
assert_eq!(optimized[1].source_statement_count, 1);
assert_eq!(optimized[2].source_statement_count, 1);
assert_eq!(optimized[3].kind, super::SqlFileImportStatementKind::Skip);
}
#[test]
fn split_batches_by_go() {
assert_eq!(super::split_sql_batches("SELECT 1\nGO\nSELECT 2"), vec!["SELECT 1", "SELECT 2"]);
}
#[test]
fn split_batches_go_case_insensitive() {
assert_eq!(
super::split_sql_batches("SELECT 1\ngo\nSELECT 2\nGo\nSELECT 3"),
vec!["SELECT 1", "SELECT 2", "SELECT 3"]
);
}
#[test]
fn split_batches_go_with_surrounding_whitespace() {
assert_eq!(super::split_sql_batches("SELECT 1\n GO \nSELECT 2"), vec!["SELECT 1", "SELECT 2"]);
}
#[test]
fn split_batches_no_go_returns_whole() {
assert_eq!(
super::split_sql_batches("DECLARE @x INT = 1;\nSELECT @x;"),
vec!["DECLARE @x INT = 1;\nSELECT @x;"]
);
}
#[test]
fn split_batches_skips_empty_batches() {
assert_eq!(super::split_sql_batches("SELECT 1\nGO\n\nGO\nSELECT 2"), vec!["SELECT 1", "SELECT 2"]);
}
#[test]
fn split_batches_trailing_go() {
assert_eq!(super::split_sql_batches("SELECT 1\nGO"), vec!["SELECT 1"]);
}
#[test]
fn split_batches_keeps_go_inside_multiline_string() {
assert_eq!(
super::split_sql_batches("SELECT 'first line\nGO\nlast line';\nGO\nSELECT 2"),
vec!["SELECT 'first line\nGO\nlast line';", "SELECT 2"]
);
}
#[test]
fn split_batches_keeps_go_inside_block_comment() {
assert_eq!(
super::split_sql_batches("SELECT 1;\n/*\nGO\n*/\nSELECT 2;\nGO\nSELECT 3;"),
vec!["SELECT 1;\n/*\nGO\n*/\nSELECT 2;", "SELECT 3;"]
);
}
// --- DELIMITER support ---
#[test]
fn delimiter_basic_procedure() {
let sql = "\
DELIMITER //
CREATE PROCEDURE foo()
BEGIN
SELECT 1;
SELECT 2;
END //
DELIMITER ;
SELECT 3;";
assert_eq!(
super::split_sql_statements(sql),
vec!["CREATE PROCEDURE foo()\nBEGIN\n SELECT 1;\n SELECT 2;\nEND", "SELECT 3",]
);
}
#[test]
fn delimiter_no_trailing_newline() {
let sql = "DELIMITER //\nSELECT 1//";
assert_eq!(super::split_sql_statements(sql), vec!["SELECT 1"]);
}
#[test]
fn delimiter_no_space_before_delim() {
let sql = "DELIMITER //\nCREATE PROCEDURE foo() BEGIN SELECT 1; END//\nDELIMITER ;";
assert_eq!(super::split_sql_statements(sql), vec!["CREATE PROCEDURE foo() BEGIN SELECT 1; END"]);
}
#[test]
fn delimiter_reset_without_whitespace() {
let sql = "DELIMITER //\nCREATE PROCEDURE foo() BEGIN SELECT 1; END//\nDELIMITER;\nCALL foo();";
assert_eq!(super::split_sql_statements(sql), vec!["CREATE PROCEDURE foo() BEGIN SELECT 1; END", "CALL foo()"]);
}
#[test]
fn delimiter_case_insensitive() {
let sql = "delimiter //\nSELECT 1//\ndelimiter ;\nSELECT 2;";
assert_eq!(super::split_sql_statements(sql), vec!["SELECT 1", "SELECT 2"]);
}
#[test]
fn delimiter_double_dollar() {
let sql = "DELIMITER $$\nCREATE FUNCTION f() RETURNS INT BEGIN RETURN 1; END $$\nDELIMITER ;";
assert_eq!(super::split_sql_statements(sql), vec!["CREATE FUNCTION f() RETURNS INT BEGIN RETURN 1; END"]);
}
#[test]
fn delimiter_semicolons_preserved_inside_body() {
let sql = "\
DELIMITER //
CREATE TRIGGER t BEFORE INSERT ON tbl FOR EACH ROW
BEGIN
SET NEW.a = 1;
SET NEW.b = 2;
END //
DELIMITER ;";
let stmts = super::split_sql_statements(sql);
assert_eq!(stmts.len(), 1);
assert!(stmts[0].contains("SET NEW.a = 1;\n SET NEW.b = 2;"));
}
#[test]
fn delimiter_multiple_statements() {
let sql = "\
DELIMITER //
CREATE PROCEDURE p1() BEGIN SELECT 1; END //
CREATE PROCEDURE p2() BEGIN SELECT 2; END //
DELIMITER ;";
assert_eq!(
super::split_sql_statements(sql),
vec!["CREATE PROCEDURE p1() BEGIN SELECT 1; END", "CREATE PROCEDURE p2() BEGIN SELECT 2; END",]
);
}
#[test]
fn delimiter_after_comment_with_chinese() {
let sql = "\
-- 判断字段是否存在
DELIMITER $$
DROP FUNCTION IF EXISTS isFieldExisting $$
CREATE FUNCTION isFieldExisting(s VARCHAR(100), t VARCHAR(100), f VARCHAR(100))
RETURNS INT
RETURN (SELECT COUNT(COLUMN_NAME)
FROM INFORMATION_SCHEMA.columns
WHERE TABLE_SCHEMA = s
AND TABLE_NAME = t
AND COLUMN_NAME = f)$$
DELIMITER ;";
let stmts = super::split_sql_statements(sql);
assert_eq!(stmts.len(), 2);
assert!(stmts[0].starts_with("DROP FUNCTION"));
assert!(stmts[1].starts_with("CREATE FUNCTION"));
}
#[test]
fn delimiter_after_ascii_comment() {
let sql = "\
-- check field existence
DELIMITER $$
SELECT 1 $$
DELIMITER ;";
assert_eq!(super::split_sql_statements(sql), vec!["SELECT 1"]);
}
#[test]
fn delimiter_after_statement() {
let sql = "\
SELECT 1;
DELIMITER $$
SELECT 2 $$
DELIMITER ;";
assert_eq!(super::split_sql_statements(sql), vec!["SELECT 1", "SELECT 2"]);
}
#[test]
fn mysql_routine_without_delimiter_keeps_body_together() {
let sql = "\
CREATE PROCEDURE p()
BEGIN
SELECT 1;
SELECT 2;
END;
SELECT 3;";
assert_eq!(
split_sql_statements_for_database(sql, DatabaseType::Mysql),
vec!["CREATE PROCEDURE p()\nBEGIN\n SELECT 1;\n SELECT 2;\nEND", "SELECT 3"]
);
}
#[test]
fn mysql_routine_without_delimiter_handles_nested_end_suffixes() {
let sql = "\
CREATE PROCEDURE p()
BEGIN
IF 1 = 1 THEN
SELECT 'ok';
END IF;
END;
SELECT 2;";
assert_eq!(
split_sql_statements_for_database(sql, DatabaseType::Mysql),
vec!["CREATE PROCEDURE p()\nBEGIN\n IF 1 = 1 THEN\n SELECT 'ok';\n END IF;\nEND", "SELECT 2"]
);
}
#[test]
fn mysql_routine_without_delimiter_handles_loop_end_suffixes() {
let sql = "\
CREATE PROCEDURE p_loop()
BEGIN
WHILE 1 = 0 DO
SELECT 'while; still body';
END WHILE;
REPEAT
SELECT 'repeat; still body';
UNTIL 1 = 1 END REPEAT;
END;
SELECT 2;";
assert_eq!(
split_sql_statements_for_database(sql, DatabaseType::Mysql),
vec![
"CREATE PROCEDURE p_loop()\nBEGIN\n WHILE 1 = 0 DO\n SELECT 'while; still body';\n END WHILE;\n REPEAT\n SELECT 'repeat; still body';\n UNTIL 1 = 1 END REPEAT;\nEND",
"SELECT 2",
]
);
}
#[test]
fn mysql_routine_without_delimiter_handles_case_endings() {
let expression = "\
CREATE PROCEDURE p_case()
BEGIN
INSERT INTO audit_log (status_text)
SELECT CASE WHEN active = 1 THEN 'active' ELSE 'inactive' END;
DELETE FROM stale_rows WHERE expires_at < NOW();
END;
SELECT 2;";
assert_eq!(
split_sql_statements_for_database(expression, DatabaseType::Mysql),
vec![
"CREATE PROCEDURE p_case()\nBEGIN\n INSERT INTO audit_log (status_text)\n SELECT CASE WHEN active = 1 THEN 'active' ELSE 'inactive' END;\n DELETE FROM stale_rows WHERE expires_at < NOW();\nEND",
"SELECT 2",
]
);
let statement = "\
CREATE PROCEDURE p_case_statement()
BEGIN
CASE WHEN active = 1 THEN SELECT 1; ELSE SELECT 0; END CASE;
DELETE FROM stale_rows WHERE expires_at < NOW();
END;
SELECT 2;";
assert_eq!(
split_sql_statements_for_database(statement, DatabaseType::Mysql),
vec![
"CREATE PROCEDURE p_case_statement()\nBEGIN\n CASE WHEN active = 1 THEN SELECT 1; ELSE SELECT 0; END CASE;\n DELETE FROM stale_rows WHERE expires_at < NOW();\nEND",
"SELECT 2",
]
);
}
#[test]
fn mysql_routine_without_delimiter_closes_nested_case_begin_blocks_in_order() {
let sql = "\
CREATE PROCEDURE p_nested_case()
BEGIN
CASE WHEN active = 1 THEN BEGIN SELECT 1; END; ELSE SELECT 0; END CASE;
DELETE FROM stale_rows WHERE expires_at < NOW();
END;
SELECT 2;";
assert_eq!(
split_sql_statements_for_database(sql, DatabaseType::Mysql),
vec![
"CREATE PROCEDURE p_nested_case()\nBEGIN\n CASE WHEN active = 1 THEN BEGIN SELECT 1; END; ELSE SELECT 0; END CASE;\n DELETE FROM stale_rows WHERE expires_at < NOW();\nEND",
"SELECT 2",
]
);
}
#[test]
fn mysql_regular_begin_transaction_still_splits_without_delimiter() {
assert_eq!(
split_sql_statements_for_database("BEGIN; INSERT INTO t VALUES (1); COMMIT;", DatabaseType::Mysql),
vec!["BEGIN", "INSERT INTO t VALUES (1)", "COMMIT"]
);
}
#[test]
fn sql_dialect_profiles_map_database_types_to_parser_capabilities() {
let default = SqlDialectProfile::default();
assert!(default.supports_custom_delimiter_commands);
assert!(default.supports_dollar_quoted_strings);
assert!(!default.supports_hash_line_comments);
assert!(default.supports_backslash_escaped_quotes);
assert!(!default.supports_postgres_escape_strings);
assert!(!default.supports_oracle_q_quotes);
assert!(SqlDialectProfile::for_database_type(DatabaseType::Oracle).supports_oracle_q_quotes);
assert!(SqlDialectProfile::for_database_type(DatabaseType::Dameng).supports_oracle_q_quotes);
assert!(!default.supports_nested_block_comments);
assert!(!default.supports_mysql_routine_blocks);
assert!(!default.supports_oracle_plsql_blocks);
assert!(!default.supports_oracle_style_routine_bodies);
assert!(!default.supports_hana_script_blocks);
assert!(!default.supports_firebird_routine_blocks);
assert!(!default.supports_slash_line_block_delimiter);
assert!(!default.supports_go_batch_separator);
assert!(!default.keeps_sqlserver_module_batch_at_cursor);
assert!(!default.preserves_tdsql_leading_directives);
for db_type in [DatabaseType::Postgres, DatabaseType::OpenGauss] {
let profile = SqlDialectProfile::for_database_type(db_type);
assert_eq!(profile, SqlDialectProfile::postgres_family());
assert!(!profile.supports_oracle_plsql_blocks);
assert!(profile.supports_oracle_style_routine_bodies);
assert!(profile.supports_postgres_escape_strings);
assert!(profile.supports_nested_block_comments);
assert!(!profile.supports_backslash_escaped_quotes);
assert!(profile.supports_slash_line_block_delimiter);
}
let mysql = SqlDialectProfile::for_database_type(DatabaseType::Mysql);
assert!(mysql.supports_hash_line_comments);
assert!(mysql.supports_backslash_escaped_quotes);
assert!(!mysql.supports_nested_block_comments);
assert!(mysql.supports_mysql_routine_blocks);
assert!(mysql.preserves_tdsql_leading_directives);
assert!(SqlDialectProfile::for_database_type(DatabaseType::Doris).supports_hash_line_comments);
assert!(!SqlDialectProfile::for_database_type(DatabaseType::Doris).preserves_tdsql_leading_directives);
assert!(SqlDialectProfile::for_database_type(DatabaseType::StarRocks).supports_hash_line_comments);
assert!(SqlDialectProfile::for_database_type(DatabaseType::ManticoreSearch).supports_hash_line_comments);
assert!(SqlDialectProfile::for_database_type(DatabaseType::Goldendb).supports_hash_line_comments);
assert!(SqlDialectProfile::for_database_type(DatabaseType::Gaussdb).supports_postgres_escape_strings);
assert!(SqlDialectProfile::for_database_type(DatabaseType::SqlServer).supports_nested_block_comments);
for db_type in [
DatabaseType::Vastbase,
DatabaseType::Kingbase,
DatabaseType::Highgo,
DatabaseType::Uxdb,
DatabaseType::Kwdb,
] {
let profile = SqlDialectProfile::for_database_type(db_type);
assert!(!profile.supports_backslash_escaped_quotes);
assert!(profile.supports_postgres_escape_strings);
}
let duckdb = SqlDialectProfile::for_database_type(DatabaseType::DuckDb);
assert!(!duckdb.supports_backslash_escaped_quotes);
assert!(duckdb.supports_postgres_escape_strings);
assert!(SqlDialectProfile::for_database_type(DatabaseType::ClickHouse).supports_backslash_escaped_quotes);
for db_type in [
DatabaseType::Oracle,
DatabaseType::Dameng,
DatabaseType::Yashandb,
DatabaseType::Oscar,
DatabaseType::OceanbaseOracle,
] {
let profile = SqlDialectProfile::for_database_type(db_type);
assert!(profile.supports_oracle_q_quotes);
assert!(!profile.supports_backslash_escaped_quotes);
assert!(profile.supports_oracle_plsql_blocks);
assert!(profile.supports_slash_line_block_delimiter);
assert!(!profile.supports_postgres_dollar_quoted_routines);
}
let gaussdb = SqlDialectProfile::for_database_type(DatabaseType::Gaussdb);
assert_eq!(gaussdb, SqlDialectProfile::gaussdb());
assert!(gaussdb.supports_oracle_plsql_blocks);
assert!(gaussdb.supports_slash_line_block_delimiter);
assert!(gaussdb.supports_postgres_dollar_quoted_routines);
let sql_server = SqlDialectProfile::for_database_type(DatabaseType::SqlServer);
assert_eq!(sql_server, SqlDialectProfile::sql_server());
assert!(sql_server.supports_go_batch_separator);
assert!(sql_server.keeps_sqlserver_module_batch_at_cursor);
let sap_hana = SqlDialectProfile::for_database_type(DatabaseType::SapHana);
assert_eq!(sap_hana, SqlDialectProfile::sap_hana());
assert!(sap_hana.supports_hana_script_blocks);
let firebird = SqlDialectProfile::for_database_type(DatabaseType::Firebird);
assert_eq!(firebird, SqlDialectProfile::firebird());
assert!(firebird.supports_firebird_routine_blocks);
assert!(!firebird.supports_backslash_escaped_quotes);
for db_type in [DatabaseType::Jdbc, DatabaseType::Hive] {
let profile = SqlDialectProfile::for_database_type(db_type);
assert!(profile.supports_oracle_style_routine_bodies);
assert!(!profile.supports_oracle_plsql_blocks);
assert!(!profile.supports_mysql_routine_blocks);
assert!(!profile.supports_slash_line_block_delimiter);
assert!(profile.supports_backslash_escaped_quotes);
}
}
#[test]
fn sap_hana_split_keeps_do_block_together() {
let sql = "\
DO
BEGIN
SELECT 1 AS \"Result\" FROM DUMMY;
END;
SELECT 2 FROM DUMMY;";
assert_eq!(
split_sql_statements_for_database(sql, DatabaseType::SapHana),
vec!["DO\nBEGIN\n SELECT 1 AS \"Result\" FROM DUMMY;\nEND;", "SELECT 2 FROM DUMMY"]
);
}
#[test]
fn sap_hana_split_keeps_alter_procedure_body_together() {
let sql = "ALTER PROCEDURE PUBLIC.P\nLANGUAGE SQLSCRIPT\nSQL SECURITY INVOKER\nAS\nBEGIN\n DECLARE v INT;\n SELECT 1 INTO v FROM DUMMY;\nEND;\nSELECT 2 FROM DUMMY;";
assert_eq!(
split_sql_statements_for_database(sql, DatabaseType::SapHana),
vec![
"ALTER PROCEDURE PUBLIC.P\nLANGUAGE SQLSCRIPT\nSQL SECURITY INVOKER\nAS\nBEGIN\n DECLARE v INT;\n SELECT 1 INTO v FROM DUMMY;\nEND;",
"SELECT 2 FROM DUMMY"
]
);
}
#[test]
fn sap_hana_split_keeps_nested_case_group_by_inside_procedure_together() {
let sql = "ALTER PROCEDURE PUBLIC.P\nLANGUAGE SQLSCRIPT\nSQL SECURITY INVOKER\nAS\nBEGIN\n T = SELECT 1 FROM DUMMY\n GROUP BY\n CASE WHEN KHLX='老客' THEN CASE WHEN SCYEAR='0000' THEN NULL ELSE SCYEAR END ELSE LEFT(CASE WHEN SCDAT_OLD='00000000' THEN NULL ELSE SCDAT_OLD END,4) END,\n CASE WHEN KHLX='老客' THEN CASE WHEN SCDAT='00000000' THEN NULL ELSE SCDAT END ELSE (CASE WHEN SCDAT_OLD='00000000' THEN NULL ELSE SCDAT_OLD END)END;\nEND;\nSELECT 2 FROM DUMMY;";
assert_eq!(
split_sql_statements_for_database(sql, DatabaseType::SapHana),
vec![sql[..sql.rfind("\nSELECT 2").unwrap()].to_string(), "SELECT 2 FROM DUMMY".to_string()]
);
}
#[test]
fn sap_hana_split_handles_single_line_and_lowercase_end() {
for sql in [
"DO BEGIN T = SELECT CASE WHEN 1 = 1 THEN 1 ELSE 0 END FROM DUMMY; END;",
"do begin t = select 1 from dummy; end;",
] {
assert_eq!(split_sql_statements_for_database(sql, DatabaseType::SapHana), vec![sql.to_string()]);
}
}
#[test]
fn sap_hana_plain_alter_procedure_still_flushes_at_semicolon() {
assert_eq!(
split_sql_statements_for_database(
"ALTER PROCEDURE PUBLIC.P RECOMPILE; SELECT 1 FROM DUMMY;",
DatabaseType::SapHana
),
vec!["ALTER PROCEDURE PUBLIC.P RECOMPILE", "SELECT 1 FROM DUMMY"]
);
}
#[test]
fn sap_hana_cursor_statement_keeps_nested_do_block_together() {
let sql = "\
DO
BEGIN
IF 1 = 1 THEN
SELECT CASE WHEN 1 = 1 THEN 1 ELSE 0 END AS \"Result\" FROM DUMMY;
END IF;
END;
SELECT 2 FROM DUMMY;";
let cursor = sql.find("Result").unwrap();
assert_eq!(
find_statement_at_cursor_for_database(sql, cursor, DatabaseType::SapHana),
"DO\nBEGIN\n IF 1 = 1 THEN\n SELECT CASE WHEN 1 = 1 THEN 1 ELSE 0 END AS \"Result\" FROM DUMMY;\n END IF;\nEND;"
);
}
#[test]
fn firebird_split_keeps_trigger_block_together() {
let sql = "\
CREATE OR ALTER TRIGGER \"WL_FYD_TP_AU\" FOR \"WL_FYD\" ACTIVE
AFTER UPDATE
as
begin
/* Trigger text */
IF ( COALESCE(NEW.QSTPS,0) <> COALESCE(OLD.QSTPS,0) ) THEN
BEGIN
UPDATE WL_FYD SET ZT='完成' WHERE ID= NEW.ID;
END
end;
SELECT 2 FROM RDB$DATABASE;";
assert_eq!(
split_sql_statements_for_database(sql, DatabaseType::Firebird),
vec![
"CREATE OR ALTER TRIGGER \"WL_FYD_TP_AU\" FOR \"WL_FYD\" ACTIVE\nAFTER UPDATE\nas\nbegin\n /* Trigger text */\n IF ( COALESCE(NEW.QSTPS,0) <> COALESCE(OLD.QSTPS,0) ) THEN\n BEGIN\n UPDATE WL_FYD SET ZT='完成' WHERE ID= NEW.ID;\n END\nend;",
"SELECT 2 FROM RDB$DATABASE"
]
);
}
#[test]
fn firebird_split_keeps_procedure_with_declarations_together() {
let sql = "\
CREATE OR ALTER PROCEDURE SP_GET_USER (USER_ID INT)
RETURNS (USERNAME VARCHAR(50), TOTAL_AMOUNT NUMERIC(18, 2))
AS
DECLARE VARIABLE V_TEMP NUMERIC(18, 2);
DECLARE VARIABLE V_STATUS INT;
BEGIN
SELECT FIRST 1 NAME FROM USERS WHERE ID = :USER_ID INTO :USERNAME;
V_TEMP = 100.50;
TOTAL_AMOUNT = :V_TEMP;
SUSPEND;
END;
SELECT 1 FROM RDB$DATABASE;";
assert_eq!(
split_sql_statements_for_database(sql, DatabaseType::Firebird),
vec![
"CREATE OR ALTER PROCEDURE SP_GET_USER (USER_ID INT)\nRETURNS (USERNAME VARCHAR(50), TOTAL_AMOUNT NUMERIC(18, 2))\nAS\nDECLARE VARIABLE V_TEMP NUMERIC(18, 2);\nDECLARE VARIABLE V_STATUS INT;\nBEGIN\n SELECT FIRST 1 NAME FROM USERS WHERE ID = :USER_ID INTO :USERNAME;\n V_TEMP = 100.50;\n TOTAL_AMOUNT = :V_TEMP;\n SUSPEND;\nEND;",
"SELECT 1 FROM RDB$DATABASE"
]
);
}
#[test]
fn firebird_split_keeps_execute_block_together() {
let sql = "\
EXECUTE BLOCK
RETURNS (OUT_ID INT)
AS
DECLARE VARIABLE I INT;
BEGIN
I = 0;
WHILE (I < 10) DO
BEGIN
I = I + 1;
END
OUT_ID = I;
SUSPEND;
END;
SELECT 1 FROM RDB$DATABASE;";
assert_eq!(
split_sql_statements_for_database(sql, DatabaseType::Firebird),
vec![
"EXECUTE BLOCK\nRETURNS (OUT_ID INT)\nAS\nDECLARE VARIABLE I INT;\nBEGIN\n I = 0;\n WHILE (I < 10) DO\n BEGIN\n I = I + 1;\n END\n OUT_ID = I;\n SUSPEND;\nEND;",
"SELECT 1 FROM RDB$DATABASE"
]
);
}
#[test]
fn firebird_split_keeps_routine_with_case_expression_together() {
let sql = "\
CREATE PROCEDURE P_CASE
AS
BEGIN
UPDATE T SET FLAG = CASE WHEN X > 0 THEN 1 ELSE 0 END;
END;
SELECT 1 FROM RDB$DATABASE;";
assert_eq!(
split_sql_statements_for_database(sql, DatabaseType::Firebird),
vec![
"CREATE PROCEDURE P_CASE\nAS\nBEGIN\n UPDATE T SET FLAG = CASE WHEN X > 0 THEN 1 ELSE 0 END;\nEND;",
"SELECT 1 FROM RDB$DATABASE"
]
);
}
#[test]
fn firebird_plain_alter_trigger_still_flushes_at_semicolon() {
assert_eq!(
split_sql_statements_for_database(
"ALTER TRIGGER TR_USERS INACTIVE; SELECT 1 FROM RDB$DATABASE;",
DatabaseType::Firebird
),
vec!["ALTER TRIGGER TR_USERS INACTIVE", "SELECT 1 FROM RDB$DATABASE"]
);
}
#[test]
fn firebird_cursor_statement_returns_whole_trigger() {
let sql = "\
CREATE OR ALTER TRIGGER \"WL_FYD_TP_AU\" FOR \"WL_FYD\" ACTIVE
AFTER UPDATE
as
begin
/* Trigger text */
IF ( COALESCE(NEW.QSTPS,0) <> COALESCE(OLD.QSTPS,0) ) THEN
BEGIN
UPDATE WL_FYD SET ZT='完成' WHERE ID= NEW.ID;
END
end;
SELECT 2 FROM RDB$DATABASE;";
let cursor = sql.find("UPDATE WL_FYD").unwrap();
assert_eq!(
find_statement_at_cursor_for_database(sql, cursor, DatabaseType::Firebird),
"CREATE OR ALTER TRIGGER \"WL_FYD_TP_AU\" FOR \"WL_FYD\" ACTIVE\nAFTER UPDATE\nas\nbegin\n /* Trigger text */\n IF ( COALESCE(NEW.QSTPS,0) <> COALESCE(OLD.QSTPS,0) ) THEN\n BEGIN\n UPDATE WL_FYD SET ZT='完成' WHERE ID= NEW.ID;\n END\nend;"
);
}
#[test]
fn oracle_split_keeps_anonymous_block_with_local_routines_together() {
let cases = [
"DECLARE\nPROCEDURE local_proc IS\nBEGIN\n NULL;\nEND;\n\nBEGIN\nlocal_proc;\nEND;",
"DECLARE\n v NUMBER;\n PROCEDURE p1 IS BEGIN NULL; END;\n FUNCTION f1 RETURN NUMBER IS BEGIN RETURN 1; END f1;\nBEGIN\n p1;\n v := f1;\nEND;",
"DECLARE\n PROCEDURE fwd(x NUMBER);\n PROCEDURE fwd(x NUMBER) IS BEGIN NULL; END;\nBEGIN\n fwd(1);\nEND;",
"DECLARE\n PROCEDURE outer_p IS\n PROCEDURE inner_p IS BEGIN NULL; END;\n BEGIN\n inner_p;\n END;\nBEGIN\n outer_p;\nEND;",
];
for sql in cases {
assert_eq!(split_sql_statements_for_database(sql, DatabaseType::Oracle), vec![sql.to_string()], "{sql}");
}
let block = cases[0];
let sql = format!("{block}\nSELECT 1 FROM dual;");
assert_eq!(
split_sql_statements_for_database(&sql, DatabaseType::Oracle),
vec![block.to_string(), "SELECT 1 FROM dual".to_string()]
);
}
// A stored routine declares its local subprograms right after its own IS/AS, with
// no DECLARE keyword, so their END must not finish the routine.
#[test]
fn oracle_split_keeps_stored_routines_with_local_subprograms_together() {
let cases = [
"CREATE OR REPLACE EDITIONABLE PROCEDURE \"A\".\"P\" AS\n PROCEDURE inner_p IS BEGIN NULL; END;\nBEGIN\n inner_p;\nEND;",
"CREATE PROCEDURE p(x NUMBER) IS\n v NUMBER;\n PROCEDURE inner_p IS\n BEGIN\n NULL;\n END inner_p;\nBEGIN\n inner_p;\nEND p;",
"CREATE OR REPLACE FUNCTION f RETURN NUMBER AS\n FUNCTION g RETURN NUMBER IS BEGIN RETURN 1; END g;\nBEGIN\n RETURN g;\nEND;",
"CREATE OR REPLACE PROCEDURE p AS\n PROCEDURE fwd(x NUMBER);\n PROCEDURE fwd(x NUMBER) IS BEGIN NULL; END;\nBEGIN\n fwd(1);\nEND;",
"CREATE OR REPLACE PROCEDURE p AS\n CURSOR c IS SELECT 1 FROM dual;\nBEGIN\n NULL;\nEND;",
];
for routine in cases {
for database in [DatabaseType::Oracle, DatabaseType::Dameng, DatabaseType::Xugu] {
assert_eq!(
split_sql_statements_for_database(&format!("{routine}\nSELECT 1 FROM dual;"), database),
vec![routine.to_string(), "SELECT 1 FROM dual".to_string()],
"{database:?}: {routine}"
);
}
}
}
#[test]
fn oracle_split_ends_call_trigger_before_following_ddl() {
let trigger = "CREATE OR REPLACE TRIGGER \"S\".\"TR\" BEFORE INSERT ON \"S\".\"T\" FOR EACH ROW CALL \"S\".\"P\"(:NEW.ID, 'a;b');";
for database in [DatabaseType::Oracle, DatabaseType::OceanbaseOracle] {
for separator in ["\n", "\n/\n"] {
let sql = format!("{trigger}{separator}ALTER TRIGGER \"S\".\"TR\" DISABLE; COMMENT ON TABLE \"S\".\"T\" IS 'unchanged';");
assert_eq!(
split_sql_statements_for_database(&sql, database),
vec![
trigger.to_string(),
"ALTER TRIGGER \"S\".\"TR\" DISABLE".to_string(),
"COMMENT ON TABLE \"S\".\"T\" IS 'unchanged'".to_string()
]
);
}
}
}
#[test]
fn oracle_split_keeps_compound_trigger_together() {
let trigger = "\
CREATE OR REPLACE EDITIONABLE TRIGGER \"A\".\"T\" FOR INSERT OR UPDATE ON \"A\".\"X\" COMPOUND TRIGGER
total NUMBER := 0;
BEFORE EACH ROW IS
BEGIN
IF :NEW.ID IS NULL THEN total := total + 1; END IF;
END BEFORE EACH ROW;
AFTER STATEMENT IS
BEGIN
NULL;
END AFTER STATEMENT;
END \"T\";";
assert_eq!(
split_sql_statements_for_database(
&format!("{trigger}\nALTER TRIGGER \"A\".\"T\" ENABLE;"),
DatabaseType::Oracle
),
vec![trigger.to_string(), "ALTER TRIGGER \"A\".\"T\" ENABLE".to_string()]
);
// A simple trigger still ends at its own END.
let simple = "CREATE TRIGGER t BEFORE INSERT ON x FOR EACH ROW BEGIN NULL; END;";
assert_eq!(
split_sql_statements_for_database(&format!("{simple}\nSELECT 1 FROM dual;"), DatabaseType::Oracle),
vec![simple.to_string(), "SELECT 1 FROM dual".to_string()]
);
}
#[test]
fn oracle_split_keeps_compound_trigger_sections_in_one_execution_unit() {
let trigger = "CREATE OR REPLACE TRIGGER \"S\".\"tr\" FOR INSERT ON \"S\".\"t\" COMPOUND TRIGGER\nBEFORE STATEMENT IS BEGIN NULL; END BEFORE STATEMENT;\nAFTER EACH ROW IS BEGIN :NEW.x := q'[a;b]'; END AFTER EACH ROW;\nEND tr;";
let state = "ALTER TRIGGER \"S\".\"tr\" DISABLE";
for database in [DatabaseType::Oracle, DatabaseType::OceanbaseOracle] {
assert_eq!(split_sql_statements_for_database(trigger, database), vec![trigger.to_string()]);
assert_eq!(
split_sql_statements_for_database(
&format!("{trigger}\n/\n{state};\nCREATE TABLE t2 (id NUMBER);"),
database
),
vec![trigger.to_string(), state.to_string(), "CREATE TABLE t2 (id NUMBER)".to_string()]
);
let execution_units = [trigger.to_string(), format!("{state};")];
let split_units: Vec<_> =
execution_units.iter().flat_map(|unit| split_sql_statements_for_database(unit, database)).collect();
assert_eq!(split_units, vec![trigger.to_string(), state.to_string()]);
}
}
#[test]
fn oracle_like_split_keeps_anonymous_plsql_block_together() {
let sql = "\
DECLARE
V_EXISTS_FLAG NUMBER;
BEGIN
SELECT COUNT(1)
INTO V_EXISTS_FLAG
FROM SCHEMA_NAME.TABLE_NAME
WHERE UNIQUE_KEY_COLUMN = 'BUSINESS_UNIQUE_VALUE';
IF V_EXISTS_FLAG = 0 THEN
INSERT INTO SCHEMA_NAME.TABLE_NAME (UNIQUE_KEY_COLUMN, COLUMN_NAME_1)
VALUES ('BUSINESS_UNIQUE_VALUE', 'BUSINESS_VALUE_1');
END IF;
END;";
assert_eq!(split_sql_statements_for_database(sql, DatabaseType::Oracle), vec![sql.to_string()]);
assert_eq!(split_sql_statements_for_database(sql, DatabaseType::Dameng), vec![sql.to_string()]);
assert_eq!(split_sql_statements_for_database(sql, DatabaseType::Gaussdb), vec![sql.to_string()]);
assert_eq!(split_sql_statements_for_database(sql, DatabaseType::Xugu), vec![sql.to_string()]);
}
#[test]
fn oracle_like_split_keeps_quoted_identifier_block_together_per_issue_10434() {
// `BEGIN "SCHEMA"."PROCEDURE"(); END;` is what the DM (Dameng) routine dialog
// generates. The double-quoted name used to be dropped from the token stream, so the
// block tokenized exactly like a transaction `BEGIN;` and was cut at its inner
// semicolon — the server then rejected the fragment with 42000 + vendorCode -2007
// ("第 2 行, 第 32 列[]附近出现错误: 语法分析出错").
let sql = "BEGIN\n \"DLJPLAT\".\"用户表脱敏\"();\nEND;";
for db_type in [DatabaseType::Dameng, DatabaseType::Oracle, DatabaseType::Xugu, DatabaseType::Gaussdb] {
assert_eq!(split_sql_statements_for_database(sql, db_type), vec![sql.to_string()], "{db_type:?}");
}
// A quoted identifier that spells a block keyword is a name, not a block boundary.
let keyword_named = "BEGIN\n INSERT INTO \"END\" VALUES (1);\nEND;";
assert_eq!(
split_sql_statements_for_database(keyword_named, DatabaseType::Dameng),
vec![keyword_named.to_string()]
);
// `BEGIN;` stays a transaction statement and still splits at its own terminator.
assert_eq!(
split_sql_statements_for_database("BEGIN; INSERT INTO t VALUES (1); COMMIT;", DatabaseType::Dameng),
vec!["BEGIN", "INSERT INTO t VALUES (1)", "COMMIT"]
);
}
#[test]
fn oracle_like_split_keeps_with_function_declaration_and_main_query_together_per_issue_11275() {
// DM8 accepts Oracle's `WITH [FUNCTION ...] <main query>` form and executes it as ONE
// statement (the reporter's own client returned f = 1). Splitting at the first
// declaration `;` sent the truncated `WITH FUNCTION ... v_start TIMESTAMP := SYSTIMESTAMP`
// fragment, and DM rejected it with 42000 / vendorCode -2007
// ("第 2 行, 第 39 列[]附近出现错误: 语法分析出错").
let sql = "\
WITH FUNCTION f RETURN NUMBER IS
v_start TIMESTAMP := SYSTIMESTAMP;
v_delay_seconds NUMBER := MOD(TO_NUMBER(TO_CHAR(SYSDATE, 'SSSSS')), 10) + 1;
BEGIN
WHILE (SYSTIMESTAMP - v_start) < (v_delay_seconds * INTERVAL '1' SECOND) LOOP
NULL;
END LOOP;
RETURN 1;
END;
SELECT f FROM dual;";
for db_type in [DatabaseType::Dameng, DatabaseType::Oracle, DatabaseType::Gaussdb, DatabaseType::Xugu] {
assert_eq!(split_sql_statements_for_database(sql, db_type), vec![sql.to_string()], "{db_type:?}");
}
// The statement after the WITH query still splits at its own terminator.
let script = format!("{sql}\nSELECT 2 FROM dual;");
assert_eq!(
split_sql_statements_for_database(&script, DatabaseType::Dameng),
vec![sql.to_string(), "SELECT 2 FROM dual".to_string()]
);
// Multiple declarations stay in one statement until the main query's terminator.
let two_declarations = "\
WITH FUNCTION f1 RETURN NUMBER IS
BEGIN
RETURN 1;
END;
FUNCTION f2 RETURN NUMBER IS
v NUMBER := 0;
BEGIN
v := 2;
RETURN v;
END;
SELECT f1 + f2 FROM dual;";
assert_eq!(
split_sql_statements_for_database(two_declarations, DatabaseType::Dameng),
vec![two_declarations.to_string()]
);
// A plain CTE without PL/SQL declarations is ordinary SQL: the trailing `;` stays a
// delimiter and the statement is emitted without it.
let cte = "WITH t AS (SELECT 1 FROM dual) SELECT * FROM t;";
assert_eq!(
split_sql_statements_for_database(cte, DatabaseType::Dameng),
vec!["WITH t AS (SELECT 1 FROM dual) SELECT * FROM t".to_string()]
);
}
#[test]
fn argo_split_keeps_create_procedure_body_together() {
// ArgoDB (Transwarp) accepts PL/SQL-style procedure definitions whose body contains
// semicolons. The statement splitter must keep those semicolons inside the body rather
// than emitting one fragment per statement — otherwise the agent sends only the first
// `CREATE ... IS BEGIN INSERT INTO ...` fragment and the server responds with
// 42000 + vendorCode 1101 (see screenshots from the 2026-09-01 session).
let sql = "\
CREATE OR REPLACE PROCEDURE SP_ETL_LOG
(
II_DATDATE IN INT,
IV_SCHEMA_NAME IN STRING,
IV_PROCEDURE_NAME IN STRING,
II_STEP_ID IN INT,
IV_STEP_DESC IN STRING,
II_STEP_FLAG IN INT,
II_START_TIME IN TIMESTAMP
)
/****************************************
@AUTHOR:xiangxu
@CREATE-DATE:2015-09-06
@DESCRIPTION:处理执行信息插入日志表(ETL_LOG)
@MODIFICATION HISTORY:
#0.20150906-xiangxu-处理执行信息插入日志表
*****************************************/
IS
BEGIN
INSERT INTO dws.ETL_LOG
(
DATA_DATE,
SCHEMA_NAME,
PROCEDURE_NAME,
STEP_ID,
STEP_DESC,
STEP_FLAG,
START_TIME
)
VALUES
(
II_DATDATE,
IV_SCHEMA_NAME,
IV_PROCEDURE_NAME,
II_STEP_ID,
IV_STEP_DESC,
II_STEP_FLAG,
II_START_TIME
);
END;";
assert_eq!(split_sql_statements_for_database(sql, DatabaseType::Argo), vec![sql.to_string()]);
assert_eq!(split_sql_statements_for_database(sql, DatabaseType::Transwarp), vec![sql.to_string()]);
assert_eq!(split_sql_statements_for_database(sql, DatabaseType::Jdbc), vec![sql.to_string()]);
assert_eq!(split_sql_statements_for_database(sql, DatabaseType::Hive), vec![sql.to_string()]);
}
#[test]
fn jdbc_and_hive_keep_procedure_statement_ranges_intact() {
let sql = "\
CREATE OR REPLACE PROCEDURE pro_cbs_check_his(is_date IN VARCHAR(8))
IS
BEGIN
DELETE FROM transwarp.tmp_cbs_t00_corp_cust_his
WHERE data_date = is_date;
INSERT INTO transwarp.tmp_cbs_t00_corp_cust_his
SELECT * FROM transwarp.cbs_t00_corp_cust_his
WHERE data_date = is_date;
COMMIT;
END;";
let sql_as = "\
CREATE OR REPLACE PROCEDURE pro_cbs_check_his(is_date IN VARCHAR(8))
AS
BEGIN
DELETE FROM transwarp.tmp_cbs_t00_corp_cust_his
WHERE data_date = is_date;
COMMIT;
END;";
for db_type in [DatabaseType::Jdbc, DatabaseType::Hive, DatabaseType::Argo, DatabaseType::Transwarp] {
let statements = split_sql_statements_for_database(sql, db_type);
assert_eq!(statements, vec![sql.to_string()], "failed for {db_type:?}");
let statements_as = split_sql_statements_for_database(sql_as, db_type);
assert_eq!(statements_as, vec![sql_as.to_string()], "failed for {db_type:?}");
}
}
#[test]
fn jdbc_and_hive_split_hiveql_functions_with_string_body() {
let sql_basic = "\
CREATE FUNCTION my_udf AS 'com.example.Mask';
SELECT mask_fn(x) FROM t;";
let sql_if_not_exists = "\
CREATE FUNCTION IF NOT EXISTS my_udf AS 'com.example.Mask';
SELECT mask_fn(x) FROM t;";
let sql_temporary = "\
CREATE TEMPORARY FUNCTION my_udf AS 'com.example.Mask';
SELECT mask_fn(x) FROM t;";
for db_type in [DatabaseType::Jdbc, DatabaseType::Hive] {
assert_eq!(
split_sql_statements_for_database(sql_basic, db_type),
vec![
"CREATE FUNCTION my_udf AS 'com.example.Mask'".to_string(),
"SELECT mask_fn(x) FROM t".to_string(),
],
"failed for {db_type:?}"
);
assert_eq!(
split_sql_statements_for_database(sql_if_not_exists, db_type),
vec![
"CREATE FUNCTION IF NOT EXISTS my_udf AS 'com.example.Mask'".to_string(),
"SELECT mask_fn(x) FROM t".to_string(),
],
"failed for {db_type:?}"
);
assert_eq!(
split_sql_statements_for_database(sql_temporary, db_type),
vec![
"CREATE TEMPORARY FUNCTION my_udf AS 'com.example.Mask'".to_string(),
"SELECT mask_fn(x) FROM t".to_string(),
],
"failed for {db_type:?}"
);
}
}
#[test]
fn argo_split_keeps_procedure_with_dash_line_comments() {
// Mirrors the SP_ETL_LOG definition in the 2026-09-01 screenshot where each
// parameter row carries a trailing `--中文` line comment. The PL/SQL tokenizer must
// still recognise the procedure as a block; otherwise the splitter emits the
// parameter list, "COMMIT", and "END" as three separate statements (the failure
// mode shown in the 21:22:59 screenshot).
let sql = "\
CREATE OR REPLACE PROCEDURE SP_ETL_LOG
(
II_DATDATE IN INT, --数据日期
IV_SCHEMA_NAME IN STRING, --模式名
IV_PROCEDURE_NAME IN STRING, --存储过程名称
II_STEP_ID IN INT, --任务号
IV_STEP_DESC IN STRING, --任务描述
II_STEP_FLAG IN INT, --执行状态
II_START_TIME IN TIMESTAMP --起始时间
);
END;";
let statements = split_sql_statements_for_database(sql, DatabaseType::Argo);
assert_eq!(statements, vec![sql.to_string()], "splitter produced: {statements:#?}");
}
#[test]
fn argo_split_handles_full_user_procedure_with_block_comment_and_at_meta() {
// Mirrors the EXACT procedure text shown in the 2026-09-01 21:22:59 screenshot:
// - parameter list with trailing `--中文` comments
// - block comment `/********...********/` containing `@AUTHOR:` and `#0.20150906-...`
// which are NOT standard SQL comment markers (ArgoDB/Hive ignores them, but the
// splitter must still treat them as part of the surrounding block comment)
// - `IS BEGIN INSERT INTO ... ; END;`
// The result panel showed three split fragments; if this test passes, the splitter
// is fine and the issue is on the DBX.app side.
let sql = "CREATE OR REPLACE PROCEDURE SP_ETL_LOG\n\
(\n\
II_DATDATE IN INT, --数据日期\n\
IV_SCHEMA_NAME IN STRING, --模式名\n\
IV_PROCEDURE_NAME IN STRING, --存储过程名称\n\
II_STEP_ID IN INT, --任务号\n\
IV_STEP_DESC IN STRING, --任务描述\n\
II_STEP_FLAG IN INT, --执行状态\n\
II_START_TIME IN TIMESTAMP --起始时间\n\
)\n\
/****************************************\n\
@AUTHOR:xiangxu\n\
@CREATE-DATE:2015-09-06\n\
@DESCRIPTION:处理执行信息插入日志表(ETL_LOG)\n\
@MODIFICATION HISTORY:\n\
#0.20150906-xiangxu-处理执行信息插入日志表\n\
*****************************************/\n\
IS\n\
BEGIN INSERT INTO dws.ETL_LOG\n\
(\n\
DATA_DATE, --数据日期\n\
SCHEMA_NAME, --模式名\n\
PROCEDURE_NAME, --存储过程名称\n\
STEP_ID, --任务号\n\
STEP_DESC, --任务描述\n\
STEP_FLAG, --执行状态\n\
START_TIME --起始时间\n\
)\n\
;\n\
END;";
let statements = split_sql_statements_for_database(sql, DatabaseType::Argo);
assert_eq!(statements.len(), 1, "splitter produced {statements:#?}");
assert_eq!(statements[0], sql);
}
#[test]
fn argo_split_handles_minimal_user_procedure_no_body_semicolon() {
// Mirrors the user's 2026-09-02 10:09 procedure: simple CREATE OR REPLACE PROCEDURE
// body with no semicolon after INSERT INTO ... SELECT, only after the SELECT expression.
let sql = "CREATE OR REPLACE PROCEDURE SP_TEST_PART_A\n\
(\n\
II_DATADATE IN INT\n\
)\n\
IS\n\
I_DATADATE INT;\n\
BEGIN\n\
I_DATADATE := II_DATADATE;\n\
INSERT INTO dws.TEST_PART_PROC\n\
SELECT 1, 'testA1', I_DATADATE;\n\
END;";
let statements = split_sql_statements_for_database(sql, DatabaseType::Argo);
assert_eq!(statements.len(), 1, "splitter produced {statements:#?}");
assert_eq!(statements[0], sql);
}
#[test]
fn gaussdb_split_ignores_psql_controls_before_anonymous_block_per_issue_6468() {
let block = "\
/* 将逻辑放到匿名块中,便于捕获异常 pl/pgsql */
DECLARE
v_date_str VARCHAR(20) := '20240824';
v_date_dt DATE := TO_DATE(v_date_str, 'YYYYMMDD');
v_eff_num NUMERIC := 0;
v_start_time TIMESTAMP := CURRENT_TIMESTAMP;
v_step TEXT := '';
v_job_name VARCHAR := 'F_XY_DG_IBDISCOUNT_DET';
BEGIN
RAISE NOTICE 'issue 6468';
END;";
let sql = format!("\\set ON_ERROR_STOP on\n\\set VERBOSITY verbose\n\\timing\n\n{block}\n/");
assert_eq!(split_sql_statements_for_database(&sql, DatabaseType::Gaussdb), vec![block]);
let cursor = sql[..sql.find("RAISE NOTICE").unwrap()].encode_utf16().count();
let executable_block = &block[block.find("DECLARE").unwrap()..];
assert_eq!(find_statement_at_cursor_for_database(&sql, cursor, DatabaseType::Gaussdb), executable_block);
}
#[test]
fn gaussdb_streaming_split_ignores_chunked_psql_controls() {
let mut splitter =
SqlStatementSplitter::with_options(SqlParsingOptions::for_database_type(DatabaseType::Gaussdb));
assert!(splitter.push_chunk("\\set ON_ERROR_").is_empty());
assert!(splitter.push_chunk("STOP on\n\\tim").is_empty());
assert!(splitter.stop_on_error());
assert_eq!(
splitter.push_chunk("ing\nDECLARE\n value NUMBER := 1;\nBEGIN\n value := value + 1;\nEND;\n/\n"),
vec!["DECLARE\n value NUMBER := 1;\nBEGIN\n value := value + 1;\nEND;"]
);
assert!(splitter.finish().is_empty());
}
#[test]
fn gaussdb_streaming_split_tracks_on_error_stop_by_statement_position() {
let mut splitter =
SqlStatementSplitter::with_options(SqlParsingOptions::for_database_type(DatabaseType::Gaussdb));
let mut statements = splitter.push_chunk_with_control("SELECT 0;\nSELECT\n\\set ON_ERROR_");
statements.extend(splitter.push_chunk_with_control("STOP on\n 1;\nSELECT 2;"));
statements.extend(splitter.finish_with_control());
assert_eq!(
statements,
vec![
super::SqlStatementWithControl { sql: "SELECT 0".to_string(), stop_on_error: false },
super::SqlStatementWithControl { sql: "SELECT\n 1".to_string(), stop_on_error: true },
super::SqlStatementWithControl { sql: "SELECT 2".to_string(), stop_on_error: true },
]
);
}
#[test]
fn gaussdb_psql_control_does_not_submit_unfinished_query_buffer() {
let sql = "SELECT\n\\timing\n 1;\nSELECT 2;";
let statements = split_sql_statements_for_database(sql, DatabaseType::Gaussdb);
assert_eq!(statements.len(), 2);
assert_eq!(statements[0].split_whitespace().collect::<Vec<_>>(), vec!["SELECT", "1"]);
assert_eq!(statements[1], "SELECT 2");
let mut splitter =
SqlStatementSplitter::with_options(SqlParsingOptions::for_database_type(DatabaseType::Gaussdb));
assert!(splitter.push_chunk("SELECT\n\\tim").is_empty());
assert_eq!(splitter.push_chunk("ing\n 1;"), vec!["SELECT\n 1"]);
}
#[test]
fn gaussdb_split_does_not_ignore_psql_variable_assignment() {
for control in ["\\set tx_date 20240824", "\\set MY_ON_ERROR_STOP on", "\\set ON_ERROR_STOP_EXTRA on"] {
let sql = format!("{control}\nDECLARE\n value TEXT := '$tx_date';\nBEGIN\n NULL;\nEND;\n/");
let statements = split_sql_statements_for_database(&sql, DatabaseType::Gaussdb);
assert!(statements[0].starts_with(control));
}
}
#[test]
fn gaussdb_ignores_only_safe_psql_controls() {
assert_eq!(
super::parse_ignorable_psql_control_command("\\set ON_ERROR_STOP on"),
Some(super::PsqlControlCommand::OnErrorStop)
);
assert_eq!(
super::parse_ignorable_psql_control_command("\\set VERBOSITY verbose"),
Some(super::PsqlControlCommand::ClientDisplay)
);
assert_eq!(
super::parse_ignorable_psql_control_command("\\timing"),
Some(super::PsqlControlCommand::ClientDisplay)
);
assert_eq!(super::parse_ignorable_psql_control_command("\\set ON_ERROR_STOP off"), None);
assert_eq!(super::parse_ignorable_psql_control_command("\\set tx_date 20240824"), None);
assert_eq!(super::parse_ignorable_psql_control_command("\\set MY_ON_ERROR_STOP on"), None);
assert_eq!(super::parse_ignorable_psql_control_command("\\set ON_ERROR_STOP_EXTRA on"), None);
assert_eq!(super::parse_ignorable_psql_control_command("\\set ON_ERROR_STOP sometimes"), None);
}
#[test]
fn gaussdb_execution_plan_preserves_on_error_stop() {
let plan = super::sql_execution_plan_for_database(
"\\set ON_ERROR_STOP on\nSELECT 1;\nSELECT 2;",
DatabaseType::Gaussdb,
);
assert!(plan.stop_on_error);
assert_eq!(plan.statements, vec!["SELECT 1", "SELECT 2"]);
}
#[test]
fn postgres_family_accepts_display_controls_and_preserves_on_error_stop() {
let sql = "\\set ON_ERROR_STOP on\r\n\\echo '开始; SELECT 99; -- output'\r\n\\timing on\n\\set VERBOSITY verbose\nSELECT 1;\n\\echo done; SELECT 99; /* output */\nSELECT 2;\n\\echo";
for db_type in [DatabaseType::Postgres, DatabaseType::OpenGauss, DatabaseType::Gaussdb] {
let plan = super::sql_execution_plan_for_database(sql, db_type);
assert_eq!(plan.statements, vec!["SELECT 1", "SELECT 2"], "{db_type:?}");
assert!(plan.stop_on_error);
// Import parsing must agree regardless of where input chunks end.
for chunk_size in 1..=sql.len() {
let mut splitter = SqlStatementSplitter::with_options(SqlParsingOptions::for_database_type(db_type));
let chars = sql.chars().collect::<Vec<_>>();
let mut statements = Vec::new();
for chunk in chars.chunks(chunk_size) {
statements.extend(splitter.push_chunk_with_control(&chunk.iter().collect::<String>()));
}
statements.extend(splitter.finish_with_control());
assert_eq!(statements.iter().map(|s| s.sql.as_str()).collect::<Vec<_>>(), vec!["SELECT 1", "SELECT 2"]);
assert!(statements.iter().all(|s| s.stop_on_error));
}
}
}
#[test]
fn postgres_psql_controls_do_not_execute_unfinished_sql_buffer() {
let sql = "SELECT\n\\echo progress; -- text\n 1;\nSELECT 2;";
let plan = super::sql_execution_plan_for_database(sql, DatabaseType::Postgres);
assert_eq!(plan.statements.len(), 2);
assert_eq!(plan.statements[0].split_whitespace().collect::<Vec<_>>(), vec!["SELECT", "1"]);
assert!(!plan.stop_on_error);
let mut splitter =
SqlStatementSplitter::with_options(SqlParsingOptions::for_database_type(DatabaseType::Postgres));
assert!(splitter.push_chunk("SELECT\n\\echo progress; --").is_empty());
assert_eq!(splitter.push_chunk(" text\n 1;\nSELECT 2;"), vec!["SELECT\n 1", "SELECT 2"]);
assert!(splitter.finish().is_empty());
}
#[test]
fn postgres_psql_controls_inside_sql_literals_and_comments_are_preserved() {
for sql in [
"SELECT 'before\n\\echo literal\nafter';",
"SELECT \"before\n\\echo literal\nafter\";",
"SELECT $$before\n\\set ON_ERROR_STOP on\nafter$$;",
"SELECT $body$before\n\\echo literal\nafter$body$;",
"/* before\n\\echo comment\nafter */\nSELECT 1;",
"-- \\echo comment\nSELECT 1;",
"SELECT 'unfinished\n\\echo literal",
"SELECT $$unfinished\n\\set ON_ERROR_STOP on",
] {
let plan = super::sql_execution_plan_for_database(sql, DatabaseType::Postgres);
assert!(!plan.stop_on_error);
assert_eq!(plan.statements, vec![sql.trim_end_matches(';')]);
let mut splitter =
SqlStatementSplitter::with_options(SqlParsingOptions::for_database_type(DatabaseType::Postgres));
let mut statements = splitter.push_chunk(sql);
statements.extend(splitter.finish());
assert_eq!(statements, plan.statements);
}
}
#[test]
fn postgres_psql_display_text_preserves_editor_cursor_offsets() {
let sql = "\\echo 🚀 开始\nSELECT 1;\nSELECT 2;";
let cursor = sql[..sql.find("SELECT 2").unwrap() + "SELECT ".len()].encode_utf16().count();
assert_eq!(find_statement_at_cursor_for_database(sql, cursor, DatabaseType::Postgres), "SELECT 2");
}
#[test]
fn postgres_does_not_silently_ignore_variable_or_execution_commands() {
for command in [
"\\set target_schema private",
"\\set ON_ERROR_STOP off",
"\\set ON_ERROR_STOP on extra",
"\\echo done \\\\ SELECT 99",
"\\echo `touch should_not_run`",
"\\copy users FROM 'users.csv'",
"\\gexec",
"\\i script.sql",
"\\! echo shell",
] {
let sql = format!("{command}\nSELECT 1;");
let plan = super::sql_execution_plan_for_database(&sql, DatabaseType::Postgres);
assert!(plan.statements[0].starts_with(command), "{command}");
assert!(!plan.stop_on_error);
let mut splitter =
SqlStatementSplitter::with_options(SqlParsingOptions::for_database_type(DatabaseType::Postgres));
let mut statements = splitter.push_chunk(&sql);
statements.extend(splitter.finish());
assert!(statements[0].starts_with(command), "{command}");
}
let sql = "\\echo message\nSELECT 1;";
assert!(split_sql_statements_for_database(sql, DatabaseType::Mysql)[0].starts_with("\\echo"));
}
#[test]
fn gaussdb_split_separates_dollar_quoted_function_from_following_statements() {
let sql = "\
DROP FUNCTION IF EXISTS dbx_issue_4572_tmp_md5_uuid;
CREATE OR REPLACE FUNCTION dbx_issue_4572_tmp_md5_uuid (v_str IN TEXT) RETURNS varchar(36) LANGUAGE PLPGSQL IMMUTABLE AS $function$
DECLARE
str1 TEXT;
BEGIN
str1 := md5(v_str);
RETURN CAST(str1 AS varchar(36));
END$function$;
DROP FUNCTION IF EXISTS dbx_issue_4572_tmp_missing;";
let statements = split_sql_statements_for_database(sql, DatabaseType::Gaussdb);
assert_eq!(statements.len(), 3);
assert_eq!(statements[0], "DROP FUNCTION IF EXISTS dbx_issue_4572_tmp_md5_uuid");
assert!(statements[1].starts_with("CREATE OR REPLACE FUNCTION dbx_issue_4572_tmp_md5_uuid"));
assert!(statements[1].ends_with("END$function$"));
assert_eq!(statements[2], "DROP FUNCTION IF EXISTS dbx_issue_4572_tmp_missing");
}
#[test]
fn gaussdb_split_separates_issue_4573_procedure_from_following_statements() {
let sql = "\
CREATE OR REPLACE PROCEDURE createIndex (
dbName IN VARCHAR(32),
tableName IN VARCHAR(64),
indexInfo IN VARCHAR(64),
indexColumns IN VARCHAR(128)
) AS
DECLARE STMT TEXT;
DECLARE flag int;
BEGIN
SELECT
count(*) INTO flag
FROM
PG_CATALOG.PG_INDEXES
WHERE
schemaname = dbName
AND TABLENAME = tableName
AND INDEXNAME = indexInfo;
IF flag = 0 THEN STMT := 'CREATE INDEX ' || indexInfo || ' ON ' || dbName || '.' || tableName || '(' || indexColumns || ')';
EXECUTE STMT;
END IF;
END;
SELECT 1 AS after_procedure;
SELECT 2 AS final_statement;";
let statements = split_sql_statements_for_database(sql, DatabaseType::Gaussdb);
assert_eq!(statements.len(), 3);
assert!(statements[0].starts_with("CREATE OR REPLACE PROCEDURE createIndex"));
assert!(statements[0].ends_with("END;"));
assert_eq!(statements[1], "SELECT 1 AS after_procedure");
assert_eq!(statements[2], "SELECT 2 AS final_statement");
}
const POSTGRES_FAMILY_PROCEDURE_SCRIPT: &str = "\
CREATE OR REPLACE PROCEDURE sync_yxdyurl_and_flag()
AS DECLARE
BEGIN
update test_xm_20260913 set a = '23' where a = '1';
COMMIT;
EXCEPTION
WHEN OTHERS THEN
ROLLBACK;
RAISE;
END;
/
SELECT 1;";
const POSTGRES_FAMILY_PROCEDURE_BODY: &str = "\
CREATE OR REPLACE PROCEDURE sync_yxdyurl_and_flag()
AS DECLARE
BEGIN
update test_xm_20260913 set a = '23' where a = '1';
COMMIT;
EXCEPTION
WHEN OTHERS THEN
ROLLBACK;
RAISE;
END;";
/// Issue #8979: a GaussDB/openGauss Oracle-compatibility server reached through the plain
/// PostgreSQL connection type must not split the PL/SQL body at its inner semicolons, or the
/// server only receives an incomplete routine and answers with
/// `subprogram body is not ended correctly at end of input`.
#[test]
fn postgres_family_keeps_oracle_style_procedure_body_in_one_statement() {
for db_type in [DatabaseType::Postgres, DatabaseType::OpenGauss] {
assert_eq!(
split_sql_statements_for_database(POSTGRES_FAMILY_PROCEDURE_SCRIPT, db_type),
vec![POSTGRES_FAMILY_PROCEDURE_BODY.to_string(), "SELECT 1".to_string()],
"{db_type:?} must keep the procedure body whole and drop the slash terminator"
);
}
}
#[test]
fn postgres_family_current_statement_keeps_oracle_style_procedure_body_together() {
let sql = POSTGRES_FAMILY_PROCEDURE_SCRIPT;
let body_cursors =
[sql.find("update test_xm").unwrap(), sql.find("COMMIT").unwrap(), sql.find("RAISE").unwrap()];
for db_type in [DatabaseType::Postgres, DatabaseType::OpenGauss] {
for cursor in body_cursors {
assert_eq!(
find_statement_at_cursor_for_database(sql, cursor, db_type),
POSTGRES_FAMILY_PROCEDURE_BODY,
"{db_type:?} must resolve the cursor inside the body to the whole procedure"
);
}
assert_eq!(find_statement_at_cursor_for_database(sql, sql.len(), db_type), "SELECT 1");
}
}
/// The reported scripts end with a bare `/` on the last line without a trailing newline, so the
/// terminator has to be recognized at end of input too.
#[test]
fn postgres_family_drops_trailing_slash_line_without_trailing_newline() {
for suffix in ["", "\n"] {
let sql = format!("{POSTGRES_FAMILY_PROCEDURE_BODY}\n/{suffix}");
for db_type in [DatabaseType::Postgres, DatabaseType::OpenGauss] {
assert_eq!(
split_sql_statements_for_database(&sql, db_type),
vec![POSTGRES_FAMILY_PROCEDURE_BODY.to_string()],
"{db_type:?} must drop the trailing slash line (suffix {suffix:?})"
);
assert_eq!(
find_statement_at_cursor_for_database(&sql, sql.len(), db_type),
POSTGRES_FAMILY_PROCEDURE_BODY
);
}
}
}
#[test]
fn postgres_family_keeps_slash_terminated_routine_body_whole() {
let sql = "\
CREATE OR REPLACE FUNCTION f_pg() RETURNS int AS $$
BEGIN
RETURN 1;
END;
$$ LANGUAGE plpgsql;
/
SELECT f_pg();";
for db_type in [DatabaseType::Postgres, DatabaseType::OpenGauss] {
assert_eq!(
split_sql_statements_for_database(sql, db_type),
vec![
"CREATE OR REPLACE FUNCTION f_pg() RETURNS int AS $$
BEGIN
RETURN 1;
END;
$$ LANGUAGE plpgsql",
"SELECT f_pg()",
],
"{db_type:?} must treat the standalone slash line as a statement terminator"
);
}
}
#[test]
fn postgres_family_current_statement_keeps_dollar_quoted_routine_whole() {
let sql = "\
CREATE OR REPLACE FUNCTION f_pg() RETURNS int AS $$
BEGIN
RETURN 1;
END;
$$ LANGUAGE plpgsql;
/
SELECT f_pg();";
for db_type in [DatabaseType::Postgres, DatabaseType::OpenGauss] {
let statement = find_statement_at_cursor_for_database(sql, sql.find("RETURN 1").unwrap(), db_type);
assert!(statement.contains("LANGUAGE plpgsql"), "{db_type:?} resolved {statement:?}");
assert!(!statement.contains("SELECT f_pg()"), "{db_type:?} resolved {statement:?}");
}
}
#[test]
fn postgres_family_still_splits_ordinary_postgres_statements() {
let sql = "\
CREATE OR REPLACE FUNCTION f_pg() RETURNS int AS $$
BEGIN
RETURN 1;
END;
$$ LANGUAGE plpgsql;
CREATE FUNCTION f_sql() RETURNS int AS 'SELECT 1' LANGUAGE sql;
CREATE TRIGGER t_pg AFTER INSERT ON t_pg FOR EACH ROW EXECUTE FUNCTION f_pg();
BEGIN;
SELECT f_pg();
COMMIT;";
assert_eq!(
split_sql_statements_for_database(sql, DatabaseType::Postgres),
vec![
"CREATE OR REPLACE FUNCTION f_pg() RETURNS int AS $$
BEGIN
RETURN 1;
END;
$$ LANGUAGE plpgsql",
"CREATE FUNCTION f_sql() RETURNS int AS 'SELECT 1' LANGUAGE sql",
"CREATE TRIGGER t_pg AFTER INSERT ON t_pg FOR EACH ROW EXECUTE FUNCTION f_pg()",
"BEGIN",
"SELECT f_pg()",
"COMMIT",
]
);
}
#[test]
fn postgres_family_does_not_enable_full_oracle_plsql_blocks() {
assert!(!starts_with_oracle_style_routine_body("CREATE PROCEDURE p() LANGUAGE sql AS 'SELECT 1';"));
assert!(!starts_with_oracle_style_routine_body(
"CREATE TRIGGER t AFTER INSERT ON x FOR EACH ROW EXECUTE FUNCTION f();"
));
assert!(!starts_with_oracle_style_routine_body("DECLARE c CURSOR FOR SELECT 1;"));
assert!(!starts_with_oracle_style_routine_body(
"CREATE FUNCTION f_pg() RETURNS int AS $$\nBEGIN\n RETURN 1;\nEND;\n$$ LANGUAGE plpgsql;"
));
assert!(!starts_with_oracle_style_routine_body(
"CREATE OR REPLACE FUNCTION f_tag() RETURNS int AS $body$ BEGIN RETURN 1; END; $body$ LANGUAGE plpgsql"
));
assert!(starts_with_oracle_style_routine_body("CREATE OR REPLACE PROCEDURE p()\nAS DECLARE\nBEGIN"));
assert!(starts_with_oracle_style_routine_body("CREATE PROCEDURE p() IS BEGIN"));
assert!(starts_with_oracle_style_routine_body("CREATE OR REPLACE PACKAGE BODY pkg AS"));
}
#[test]
fn oracle_like_split_keeps_issue_2405_anonymous_plsql_block_together() {
let sql = "\
DECLARE
PRE_TRD_DATE INTEGER ;
BEGIN
SELECT 1 + 2 INTO PRE_TRD_DATE FROM DUAL;
END;";
assert_eq!(split_sql_statements_for_database(sql, DatabaseType::Oracle), vec![sql.to_string()]);
}
#[test]
fn oracle_like_current_statement_keeps_issue_2405_anonymous_plsql_block_together() {
let sql = "\
DECLARE
PRE_TRD_DATE INTEGER ;
BEGIN
SELECT 1 + 2 INTO PRE_TRD_DATE FROM DUAL;
END;";
let cursors = [
sql[..sql.find("PRE_TRD_DATE").unwrap()].encode_utf16().count(),
sql[..sql.find("SELECT 1 + 2").unwrap()].encode_utf16().count(),
sql[..sql.find("END;").unwrap()].encode_utf16().count(),
];
for cursor in cursors {
assert_eq!(find_statement_at_cursor_for_database(sql, cursor, DatabaseType::Oracle), sql);
}
}
#[test]
fn oracle_like_split_treats_slash_line_as_plsql_delimiter() {
let sql = "\
BEGIN
NULL;
END;
/
SELECT 1;";
assert_eq!(
split_sql_statements_for_database(sql, DatabaseType::Oracle),
vec!["BEGIN\n NULL;\nEND;", "SELECT 1"]
);
}
#[test]
fn oracle_like_current_statement_keeps_plsql_block_together() {
let sql = "\
DECLARE
v_exists_flag NUMBER;
BEGIN
IF v_exists_flag = 0 THEN
NULL;
END IF;
END;
/
SELECT 1;";
let cursor = sql[..sql.find("NULL").unwrap()].encode_utf16().count();
let next_cursor = sql[..sql.find("SELECT 1").unwrap()].encode_utf16().count();
assert_eq!(
find_statement_at_cursor_for_database(sql, cursor, DatabaseType::Oracle),
"DECLARE\n v_exists_flag NUMBER;\nBEGIN\n IF v_exists_flag = 0 THEN\n NULL;\n END IF;\nEND;"
);
assert_eq!(find_statement_at_cursor_for_database(sql, next_cursor, DatabaseType::Oracle), "SELECT 1");
}
#[test]
fn oracle_like_current_statement_keeps_nested_dml_plsql_block_together() {
let sql = "\
DECLARE
v_order_count NUMBER;
BEGIN
SELECT COUNT(*) INTO v_order_count
FROM \"DBX_TEST\".\"ORDERS_10K\";
IF v_order_count = 0 THEN
INSERT INTO \"DBX_TEST\".\"STORES\"
(\"ID\", \"STORE_CODE\", \"STORE_NAME\", \"CITY\", \"OPENED_AT\")
SELECT 10001, 'TEST_STORE_001', '测试门店', '上海', SYSDATE
FROM DUAL
WHERE NOT EXISTS (
SELECT 1 FROM \"DBX_TEST\".\"STORES\" WHERE \"ID\" = 10001
);
INSERT INTO \"DBX_TEST\".\"PRODUCTS\"
(\"ID\", \"SKU\", \"PRODUCT_NAME\", \"CATEGORY\", \"PRICE\")
SELECT 10001, 'TEST_SKU_001', '测试商品', '测试分类', 99.90
FROM DUAL
WHERE NOT EXISTS (
SELECT 1 FROM \"DBX_TEST\".\"PRODUCTS\" WHERE \"ID\" = 10001
);
INSERT INTO \"DBX_TEST\".\"ORDERS_10K\"
(\"ID\", \"ORDER_NO\", \"STORE_ID\", \"PRODUCT_ID\", \"CUSTOMER_NAME\", \"QUANTITY\", \"AMOUNT\", \"ORDER_STATUS\", \"CREATED_AT\")
SELECT 10001, 'TEST_ORDER_001', 10001, 10001, '测试客户', 2, 199.80, 'PAID', SYSDATE
FROM DUAL
WHERE NOT EXISTS (
SELECT 1 FROM \"DBX_TEST\".\"ORDERS_10K\" WHERE \"ORDER_NO\" = 'TEST_ORDER_001'
);
COMMIT;
END IF;
END;
/
SELECT 1;";
let expected = sql.split("\n/").next().unwrap();
let cursor = sql[..sql.find("ORDERS_10K").unwrap()].encode_utf16().count();
let next_cursor = sql[..sql.find("SELECT 1;").unwrap()].encode_utf16().count();
assert_eq!(
split_sql_statements_for_database(sql, DatabaseType::Oracle),
vec![expected.to_string(), "SELECT 1".to_string()]
);
assert_eq!(find_statement_at_cursor_for_database(sql, cursor, DatabaseType::Oracle), expected);
assert_eq!(find_statement_at_cursor_for_database(sql, next_cursor, DatabaseType::Oracle), "SELECT 1");
}
#[test]
fn oracle_like_split_keeps_transaction_begin_as_statement() {
assert_eq!(
split_sql_statements_for_database("BEGIN; INSERT INTO t VALUES (1); COMMIT;", DatabaseType::Gaussdb),
vec!["BEGIN", "INSERT INTO t VALUES (1)", "COMMIT"]
);
}
#[test]
fn oracle_like_split_keeps_create_function_together() {
let sql = "\
CREATE OR REPLACE FUNCTION number_tochar(nums VARCHAR(20))
RETURN VARCHAR(20)
AS
res VARCHAR(20);
BEGIN
RETURN '一';
END;
/
SELECT 1;";
assert_eq!(
split_sql_statements_for_database(sql, DatabaseType::Oracle),
vec!["CREATE OR REPLACE FUNCTION number_tochar(nums VARCHAR(20))\nRETURN VARCHAR(20)\nAS\n res VARCHAR(20);\nBEGIN\n RETURN '一';\nEND;", "SELECT 1"]
);
assert_eq!(
split_sql_statements_for_database(sql, DatabaseType::Xugu),
vec!["CREATE OR REPLACE FUNCTION number_tochar(nums VARCHAR(20))\nRETURN VARCHAR(20)\nAS\n res VARCHAR(20);\nBEGIN\n RETURN '一';\nEND;", "SELECT 1"]
);
assert_eq!(
split_sql_statements_for_database(sql, DatabaseType::Dameng),
vec!["CREATE OR REPLACE FUNCTION number_tochar(nums VARCHAR(20))\nRETURN VARCHAR(20)\nAS\n res VARCHAR(20);\nBEGIN\n RETURN '一';\nEND;", "SELECT 1"]
);
}
#[test]
fn oracle_like_split_keeps_create_procedure_together() {
let sql = "\
CREATE OR REPLACE PROCEDURE update_salary(p_id NUMBER, p_amount NUMBER)
AS
BEGIN
UPDATE employees SET salary = salary + p_amount WHERE id = p_id;
COMMIT;
END;
/
SELECT 1;";
assert_eq!(
split_sql_statements_for_database(sql, DatabaseType::Oracle),
vec![
"CREATE OR REPLACE PROCEDURE update_salary(p_id NUMBER, p_amount NUMBER)\nAS\nBEGIN\n UPDATE employees SET salary = salary + p_amount WHERE id = p_id;\n COMMIT;\nEND;",
"SELECT 1"
]
);
assert_eq!(
split_sql_statements_for_database(sql, DatabaseType::Xugu),
vec![
"CREATE OR REPLACE PROCEDURE update_salary(p_id NUMBER, p_amount NUMBER)\nAS\nBEGIN\n UPDATE employees SET salary = salary + p_amount WHERE id = p_id;\n COMMIT;\nEND;",
"SELECT 1"
]
);
}
#[test]
fn oracle_like_split_keeps_create_trigger_together() {
let sql = "\
CREATE TRIGGER trg_audit
BEFORE INSERT ON employees
FOR EACH ROW
BEGIN
INSERT INTO audit_log VALUES (:NEW.id, 'INSERT');
END;
/
SELECT 1;";
assert_eq!(
split_sql_statements_for_database(sql, DatabaseType::Oracle),
vec![
"CREATE TRIGGER trg_audit\nBEFORE INSERT ON employees\nFOR EACH ROW\nBEGIN\n INSERT INTO audit_log VALUES (:NEW.id, 'INSERT');\nEND;",
"SELECT 1"
]
);
assert_eq!(
split_sql_statements_for_database(sql, DatabaseType::Xugu),
vec![
"CREATE TRIGGER trg_audit\nBEFORE INSERT ON employees\nFOR EACH ROW\nBEGIN\n INSERT INTO audit_log VALUES (:NEW.id, 'INSERT');\nEND;",
"SELECT 1"
]
);
}
#[test]
fn yashandb_split_keeps_create_trigger_with_slash_delimiter_together() {
let sql = "\
CREATE TRIGGER \"TB_SC_UPDATE_TIME_TRI\"
BEFORE UPDATE ON \"TB_SC\"
FOR EACH ROW
BEGIN
IF NOT UPDATING('UPDATE_TIME') THEN
\t\t:NEW.\"UPDATE_TIME\" := CURRENT_TIMESTAMP;
END IF;
END;
/
SELECT 1;";
assert_eq!(
split_sql_statements_for_database(sql, DatabaseType::Yashandb),
vec![
"CREATE TRIGGER \"TB_SC_UPDATE_TIME_TRI\"\nBEFORE UPDATE ON \"TB_SC\"\nFOR EACH ROW\n BEGIN\n IF NOT UPDATING('UPDATE_TIME') THEN\n \t\t:NEW.\"UPDATE_TIME\" := CURRENT_TIMESTAMP;\n END IF;\nEND;",
"SELECT 1"
]
);
}
#[test]
fn oracle_like_split_keeps_create_package_together() {
let sql = "\
CREATE OR REPLACE PACKAGE pkg_utils AS
FUNCTION get_version RETURN VARCHAR2;
PROCEDURE log_message(msg VARCHAR2);
END pkg_utils;
/
SELECT 1;";
assert_eq!(
split_sql_statements_for_database(sql, DatabaseType::Oracle),
vec![
"CREATE OR REPLACE PACKAGE pkg_utils AS\n FUNCTION get_version RETURN VARCHAR2;\n PROCEDURE log_message(msg VARCHAR2);\nEND pkg_utils;",
"SELECT 1"
]
);
assert_eq!(
split_sql_statements_for_database(sql, DatabaseType::Xugu),
vec![
"CREATE OR REPLACE PACKAGE pkg_utils AS\n FUNCTION get_version RETURN VARCHAR2;\n PROCEDURE log_message(msg VARCHAR2);\nEND pkg_utils;",
"SELECT 1"
]
);
}
#[test]
fn opengauss_a_mode_split_keeps_create_package_together() {
let sql = "CREATE OR REPLACE PACKAGE pkg_utils AS\n FUNCTION get_version RETURN VARCHAR2;\n PROCEDURE log_message(msg VARCHAR2);\nEND pkg_utils;\n/\nSELECT 1;";
assert_eq!(
super::split_sql_statements_for_database_with_compatibility(sql, DatabaseType::OpenGauss, Some("A")),
vec![
"CREATE OR REPLACE PACKAGE pkg_utils AS\n FUNCTION get_version RETURN VARCHAR2;\n PROCEDURE log_message(msg VARCHAR2);\nEND pkg_utils;",
"SELECT 1"
]
);
assert_ne!(
super::split_sql_statements_for_database_with_compatibility(sql, DatabaseType::OpenGauss, Some("PG")),
super::split_sql_statements_for_database_with_compatibility(sql, DatabaseType::OpenGauss, Some("A"))
);
}
#[test]
fn xugu_split_keeps_create_package_body_together() {
let sql = "\
CREATE OR REPLACE PACKAGE BODY dbx_pkg AS
PROCEDURE ping AS
BEGIN
NULL;
END ping;
END dbx_pkg;
/
SELECT 1;";
assert_eq!(
split_sql_statements_for_database(sql, DatabaseType::Xugu),
vec![
"CREATE OR REPLACE PACKAGE BODY dbx_pkg AS\n PROCEDURE ping AS\n BEGIN\n NULL;\n END ping;\nEND dbx_pkg;",
"SELECT 1"
]
);
}
#[test]
fn xugu_split_keeps_routines_without_or_replace_together() {
let cases = [
"CREATE PROCEDURE dbx_proc_without_replace AS BEGIN NULL; END;",
"CREATE FUNCTION dbx_func_without_replace RETURN INTEGER AS BEGIN RETURN 1; END;",
"CREATE TRIGGER dbx_trigger_without_replace BEFORE INSERT ON dbx_events FOR EACH ROW BEGIN NULL; END;",
];
for statement in cases {
assert_eq!(
split_sql_statements_for_database(&format!("{statement}\nSELECT 1;"), DatabaseType::Xugu),
vec![statement.to_owned(), "SELECT 1".to_owned()],
"failed for {statement}"
);
}
}
#[test]
fn xugu_split_keeps_force_package_body_together() {
let sql = "\
CREATE OR REPLACE FORCE PACKAGE BODY dbx_pkg AS
PROCEDURE ping AS
BEGIN
NULL;
END ping;
END dbx_pkg;
SELECT 1;";
assert_eq!(
split_sql_statements_for_database(sql, DatabaseType::Xugu),
vec![
"CREATE OR REPLACE FORCE PACKAGE BODY dbx_pkg AS\n PROCEDURE ping AS\n BEGIN\n NULL;\n END ping;\nEND dbx_pkg;",
"SELECT 1"
]
);
assert_eq!(
split_sql_statements_for_database(
"CREATE OR REPLACE NOFORCE PACKAGE BODY dbx_pkg AS\n PROCEDURE ping AS\n BEGIN\n NULL;\n END ping;\nEND dbx_pkg;\nSELECT 1;",
DatabaseType::Xugu
),
vec![
"CREATE OR REPLACE NOFORCE PACKAGE BODY dbx_pkg AS\n PROCEDURE ping AS\n BEGIN\n NULL;\n END ping;\nEND dbx_pkg;",
"SELECT 1"
]
);
assert_eq!(
split_sql_statements_for_database(
"CREATE PACKAGE BODY dbx_pkg_without_replace AS\n PROCEDURE ping AS\n BEGIN\n NULL;\n END ping;\nEND dbx_pkg_without_replace;\nSELECT 1;",
DatabaseType::Xugu
),
vec![
"CREATE PACKAGE BODY dbx_pkg_without_replace AS\n PROCEDURE ping AS\n BEGIN\n NULL;\n END ping;\nEND dbx_pkg_without_replace;",
"SELECT 1"
]
);
}
#[test]
fn xugu_split_package_spec_without_slash_does_not_consume_following_sql() {
let sql = "\
CREATE OR REPLACE PACKAGE pkg_utils AS
FUNCTION get_version RETURN VARCHAR2;
PROCEDURE log_message(msg VARCHAR2);
END pkg_utils;
SELECT 1;";
assert_eq!(
split_sql_statements_for_database(sql, DatabaseType::Xugu),
vec![
"CREATE OR REPLACE PACKAGE pkg_utils AS\n FUNCTION get_version RETURN VARCHAR2;\n PROCEDURE log_message(msg VARCHAR2);\nEND pkg_utils;",
"SELECT 1"
]
);
assert_eq!(
split_sql_statements_for_database(
"CREATE OR REPLACE PACKAGE pkg_utils AS\n FUNCTION get_version RETURN VARCHAR2;\n PROCEDURE log_message(msg VARCHAR2);\nEND pkg_utils;\n/\nSELECT 1;",
DatabaseType::Xugu
),
vec![
"CREATE OR REPLACE PACKAGE pkg_utils AS\n FUNCTION get_version RETURN VARCHAR2;\n PROCEDURE log_message(msg VARCHAR2);\nEND pkg_utils;",
"SELECT 1"
]
);
assert_eq!(
split_sql_statements_for_database(
"CREATE OR REPLACE FORCE PACKAGE pkg_utils AS\n PROCEDURE ping;\nEND pkg_utils;\nSELECT 1;",
DatabaseType::Xugu
),
vec!["CREATE OR REPLACE FORCE PACKAGE pkg_utils AS\n PROCEDURE ping;\nEND pkg_utils;", "SELECT 1"]
);
assert_eq!(
split_sql_statements_for_database(
"CREATE PACKAGE pkg_utils_without_replace AS\n PROCEDURE ping;\nEND pkg_utils_without_replace;\nSELECT 1;",
DatabaseType::Xugu
),
vec![
"CREATE PACKAGE pkg_utils_without_replace AS\n PROCEDURE ping;\nEND pkg_utils_without_replace;",
"SELECT 1"
]
);
}
#[test]
fn oracle_like_split_declaration_only_package_body_before_following_dml() {
let package_body = "CREATE OR REPLACE PACKAGE BODY packageName IS\nnull;\nEND packageName;";
assert_eq!(
split_sql_statements_for_database(&format!("{package_body}\nSELECT * FROM goods;"), DatabaseType::Oracle),
vec![package_body, "SELECT * FROM goods"]
);
assert_eq!(
split_sql_statements_for_database(
&format!("{package_body}\n/\nSELECT * FROM goods;"),
DatabaseType::Oracle
),
vec![package_body, "SELECT * FROM goods"]
);
}
#[test]
fn oracle_like_split_large_package_body_in_linear_time() {
// Every inner `;` used to re-tokenize the whole body so far, so a few hundred KB
// DM8 package body spent minutes in the splitter before reaching the server (#13450).
let routines = (0..300)
.map(|i| {
format!(
" PROCEDURE p{i}(p_id INT) IS\n v_name VARCHAR2(100) := 'x;y';\n BEGIN\n IF p_id > 0 THEN\n INSERT INTO orders VALUES (p_id);\n END IF;\n FOR r IN (SELECT id FROM orders WHERE id = p_id) LOOP\n UPDATE orders SET name = v_name WHERE id = r.id;\n END LOOP;\n END p{i};\n"
)
})
.collect::<String>();
let package_body = format!("CREATE OR REPLACE PACKAGE BODY pkg_orders AS\n{routines}END pkg_orders;");
let sql = format!("{package_body}\nSELECT 1 FROM dual;");
let started = std::time::Instant::now();
let statements = split_sql_statements_for_database(&sql, DatabaseType::Dameng);
let cursor_statement = find_statement_at_cursor_for_database(&sql, 1, DatabaseType::Dameng);
let elapsed = started.elapsed();
assert_eq!(statements, vec![package_body.as_str(), "SELECT 1 FROM dual"]);
assert_eq!(cursor_statement, package_body);
assert!(elapsed < std::time::Duration::from_secs(2), "splitting took {elapsed:?}");
}
#[test]
fn xugu_split_keeps_create_type_body_together() {
let sql = "\
CREATE OR REPLACE TYPE BODY obj_t AS
MEMBER PROCEDURE ping IS
BEGIN
NULL;
END;
END;
SELECT 1;";
assert_eq!(
split_sql_statements_for_database(sql, DatabaseType::Xugu),
vec![
"CREATE OR REPLACE TYPE BODY obj_t AS\n MEMBER PROCEDURE ping IS\n BEGIN\n NULL;\n END;\nEND;",
"SELECT 1"
]
);
assert_eq!(
split_sql_statements_for_database(
"CREATE TYPE BODY obj_t_without_replace AS\n MEMBER PROCEDURE ping IS\n BEGIN\n NULL;\n END;\nEND;\nSELECT 1;",
DatabaseType::Xugu
),
vec![
"CREATE TYPE BODY obj_t_without_replace AS\n MEMBER PROCEDURE ping IS\n BEGIN\n NULL;\n END;\nEND;",
"SELECT 1"
]
);
}
#[test]
fn xugu_split_plain_create_type_object_on_semicolon() {
// Plain CREATE TYPE ends with ");" and must not wait for a nonexistent outer END.
let sql = "CREATE OR REPLACE TYPE address_t AS OBJECT (id INT);\nSELECT 1;";
assert_eq!(
split_sql_statements_for_database(sql, DatabaseType::Xugu),
vec!["CREATE OR REPLACE TYPE address_t AS OBJECT (id INT)", "SELECT 1"]
);
assert_eq!(
split_sql_statements_for_database(
"CREATE TYPE address_t_without_replace AS OBJECT (id INT);\nSELECT 1;",
DatabaseType::Xugu
),
vec!["CREATE TYPE address_t_without_replace AS OBJECT (id INT)", "SELECT 1"]
);
}
#[test]
fn oracle_like_split_keeps_case_expressions_inside_routines() {
let function = "CREATE OR REPLACE FUNCTION dbx_case_expr RETURN NUMBER AS\nBEGIN\n RETURN CASE WHEN 1 = 1 THEN CASE WHEN 2 = 2 THEN 1 ELSE 2 END ELSE 0 END;\nEND;\nSELECT 1;";
let procedure = "CREATE OR REPLACE PROCEDURE dbx_case_statement AS\nBEGIN\n CASE WHEN 1 = 1 THEN NULL; ELSE NULL; END CASE;\nEND;\nSELECT 1;";
for database in [DatabaseType::Xugu, DatabaseType::Oracle] {
assert_eq!(
split_sql_statements_for_database(function, database),
vec![
"CREATE OR REPLACE FUNCTION dbx_case_expr RETURN NUMBER AS\nBEGIN\n RETURN CASE WHEN 1 = 1 THEN CASE WHEN 2 = 2 THEN 1 ELSE 2 END ELSE 0 END;\nEND;",
"SELECT 1"
]
);
assert_eq!(
split_sql_statements_for_database(procedure, database),
vec![
"CREATE OR REPLACE PROCEDURE dbx_case_statement AS\nBEGIN\n CASE WHEN 1 = 1 THEN NULL; ELSE NULL; END CASE;\nEND;",
"SELECT 1"
]
);
}
}
#[test]
fn oracle_like_split_keeps_decode_consecutive_case_expressions_inside_procedure() {
// Punctuation between `end` and the next `case` must not look like PL/SQL `END CASE`,
// or CREATE PROCEDURE is truncated at the INSERT terminator (OceanBase OBE-00900 near insert).
let procedure = "\
CREATE OR REPLACE PROCEDURE dm.sp_demo IS
BEGIN
INSERT INTO t(c)
SELECT decode(
company_no,
'A', (case when x = 1 then 'one' else 'other' end),
'B', (case when y = 2 then 'two' else 'other' end)
) as business_type_name
FROM src;
COMMIT;
EXCEPTION
WHEN OTHERS THEN
NULL;
END sp_demo;";
for database in [DatabaseType::Oracle, DatabaseType::OceanbaseOracle] {
assert_eq!(split_sql_statements_for_database(procedure, database), vec![procedure.to_string()]);
}
}
#[test]
fn oracle_like_split_ignores_end_tokens_inside_q_quoted_strings() {
let sql = "\
BEGIN
v_text := q'[not really END;]';
NULL;
END;
SELECT 1;";
assert_eq!(
split_sql_statements_for_database(sql, DatabaseType::Oracle),
vec!["BEGIN\n v_text := q'[not really END;]';\n NULL;\nEND;", "SELECT 1"]
);
}
#[test]
fn oracle_plsql_tokenizer_falls_back_when_sqlparser_rejects_partial_literals() {
let sql = "\
BEGIN
NULL;
v_text := 'partial";
let tokens = super::oracle_plsql_tokens(sql);
assert!(tokens.iter().any(|token| token.is_word("BEGIN")));
assert!(tokens.iter().any(super::OraclePlSqlToken::is_semicolon));
}
#[test]
fn oracle_like_split_does_not_affect_create_table() {
let sql = "\
CREATE TABLE users (id NUMBER PRIMARY KEY, name VARCHAR2(100));
CREATE OR REPLACE VIEW v_users AS SELECT id, name FROM users;";
assert_eq!(
split_sql_statements_for_database(sql, DatabaseType::Oracle),
vec![
"CREATE TABLE users (id NUMBER PRIMARY KEY, name VARCHAR2(100))",
"CREATE OR REPLACE VIEW v_users AS SELECT id, name FROM users"
]
);
}
#[test]
fn oracle_like_split_does_not_treat_backslash_as_quote_escape() {
let sql = "SELECT TO_CLOB('a\\') || TO_CLOB('n};\\n}') FROM dual;\nSELECT 2 FROM dual;";
for db_type in [
DatabaseType::OceanbaseOracle,
DatabaseType::Oracle,
DatabaseType::Dameng,
DatabaseType::Yashandb,
DatabaseType::Oscar,
] {
assert_eq!(
split_sql_statements_for_database(sql, db_type),
vec!["SELECT TO_CLOB('a\\') || TO_CLOB('n};\\n}') FROM dual", "SELECT 2 FROM dual"],
"{db_type:?}"
);
}
}
#[test]
fn finds_statement_at_cursor() {
let sql = "SELECT 1; SELECT 2";
assert_eq!(super::find_statement_at_cursor(sql, 3), "SELECT 1");
assert_eq!(super::find_statement_at_cursor(sql, 12), "SELECT 2");
assert_eq!(super::find_statement_at_cursor(sql, 18), "SELECT 2");
}
#[test]
fn finds_statement_at_cursor_after_unicode_comment() {
let sql = "-- 判断字段是否存在\nSELECT 1; SELECT 2";
let cursor_byte = sql.find("SELECT 2").unwrap();
let cursor = sql[..cursor_byte].encode_utf16().count();
assert_eq!(super::find_statement_at_cursor(sql, cursor), "SELECT 2");
}
#[test]
fn finds_statement_at_cursor_after_semicolon_with_blank_line_stays_on_previous_statement() {
let sql = "SELECT 1;\n\nSELECT 2;";
let cursor = sql[..sql.find(';').unwrap() + 1].encode_utf16().count();
assert_eq!(super::find_statement_at_cursor(sql, cursor), "SELECT 1");
}
#[test]
fn finds_statement_at_cursor_after_semicolon_same_line_moves_to_next_statement() {
let sql = "SELECT 1; SELECT 2;";
let cursor = sql[..sql.find("SELECT 2").unwrap()].encode_utf16().count();
assert_eq!(super::find_statement_at_cursor(sql, cursor), "SELECT 2");
}
#[test]
fn finds_statement_at_cursor_after_double_blank_line_without_semicolon() {
let sql = "SELECT * FROM old_table\n\n\nCREATE VIEW v AS SELECT 1";
let cursor = sql[..sql.find("CREATE VIEW").unwrap()].encode_utf16().count();
assert_eq!(super::find_statement_at_cursor(sql, cursor), "CREATE VIEW v AS SELECT 1");
}
#[test]
fn keeps_create_view_statement_together_across_double_blank_line() {
let sql = "CREATE VIEW v AS\n\n\nSELECT 1";
let cursor = sql[..sql.find("SELECT 1").unwrap()].encode_utf16().count();
assert_eq!(super::find_statement_at_cursor(sql, cursor), "CREATE VIEW v AS\n\n\nSELECT 1");
}
#[test]
fn finds_statement_with_dollar_quote() {
let sql = "SELECT $$a;b$$; SELECT 2";
assert_eq!(super::find_statement_at_cursor(sql, 3), "SELECT $$a;b$$");
assert_eq!(super::find_statement_at_cursor(sql, 17), "SELECT 2");
}
#[test]
fn finds_statement_with_custom_delimiter() {
let sql = "\
DELIMITER //
CREATE PROCEDURE foo()
BEGIN
SELECT 1;
END //
DELIMITER ;
SELECT 2;";
let cursor = sql.find("SELECT 1").unwrap();
let next_cursor = sql.rfind("SELECT 2").unwrap();
assert_eq!(super::find_statement_at_cursor(sql, cursor), "CREATE PROCEDURE foo()\nBEGIN\n SELECT 1;\nEND");
assert_eq!(super::find_statement_at_cursor(sql, next_cursor), "SELECT 2");
}
#[test]
fn sqlserver_current_statement_keeps_procedure_batch_with_inner_semicolons() {
let sql = "\
CREATE OR ALTER PROCEDURE dbo.usp_demo
AS
BEGIN
SELECT 1;
SELECT 2;
END
GO
SELECT 3;";
let cursor = sql[..sql.find("SELECT 2").unwrap()].encode_utf16().count();
let next_cursor = sql[..sql.rfind("SELECT 3").unwrap()].encode_utf16().count();
assert_eq!(
super::find_statement_at_cursor_for_database(sql, cursor, DatabaseType::SqlServer),
"CREATE OR ALTER PROCEDURE dbo.usp_demo\nAS\nBEGIN\n SELECT 1;\n SELECT 2;\nEND"
);
assert_eq!(super::find_statement_at_cursor_for_database(sql, next_cursor, DatabaseType::SqlServer), "SELECT 3");
}
#[test]
fn sqlserver_current_statement_keeps_alter_procedure_batch() {
let sql = "\
ALTER PROC dbo.usp_demo
AS
BEGIN
UPDATE dbo.users SET name = name;
END";
let cursor = sql[..sql.find("UPDATE").unwrap()].encode_utf16().count();
assert_eq!(
super::find_statement_at_cursor_for_database(sql, cursor, DatabaseType::SqlServer),
"ALTER PROC dbo.usp_demo\nAS\nBEGIN\n UPDATE dbo.users SET name = name;\nEND"
);
}
#[test]
fn sqlserver_current_statement_keeps_if_else_control_flow_batch() {
let sql = "\
IF EXISTS (SELECT 1 FROM ::fn_listextendedproperty('MS_Description','USER','dbo','TABLE','Categories','COLUMN','CategoryID'))
BEGIN
EXEC sp_updateextendedproperty @name=N'MS_Description', @value=N'test', @level0type=N'USER', @level0name=N'dbo', @level1type=N'TABLE', @level1name=N'Categories', @level2type=N'COLUMN', @level2name=N'CategoryID'
END
ELSE
BEGIN
EXEC sp_addextendedproperty @name=N'MS_Description', @value=N'test', @level0type=N'USER', @level0name=N'dbo', @level1type=N'TABLE', @level1name=N'Categories', @level2type=N'COLUMN', @level2name=N'CategoryID'
END";
let update_cursor = sql[..sql.find("sp_updateextendedproperty").unwrap()].encode_utf16().count();
let add_cursor = sql[..sql.find("sp_addextendedproperty").unwrap()].encode_utf16().count();
assert_eq!(super::find_statement_at_cursor_for_database(sql, update_cursor, DatabaseType::SqlServer), sql);
assert_eq!(super::find_statement_at_cursor_for_database(sql, add_cursor, DatabaseType::SqlServer), sql);
}
// #13051: a top-level BEGIN TRY ... END TRY BEGIN CATCH ... END CATCH
// script is one batch, the same as an IF/ELSE control-flow batch.
#[test]
fn sqlserver_current_statement_keeps_top_level_try_catch_batch() {
let sql = "\
BEGIN TRY
BEGIN TRANSACTION;
UPDATE dbo.Documents SET Deleted = 0 WHERE Deleted = 1;
COMMIT TRANSACTION;
END TRY
BEGIN CATCH
IF @@TRANCOUNT > 0 ROLLBACK TRANSACTION;
THROW;
END CATCH";
let update_cursor = sql[..sql.find("UPDATE dbo.Documents").unwrap()].encode_utf16().count();
let rollback_cursor = sql[..sql.find("ROLLBACK").unwrap()].encode_utf16().count();
assert_eq!(super::find_statement_at_cursor_for_database(sql, update_cursor, DatabaseType::SqlServer), sql);
assert_eq!(super::find_statement_at_cursor_for_database(sql, rollback_cursor, DatabaseType::SqlServer), sql);
}
#[test]
fn sqlserver_current_statement_keeps_top_level_plain_begin_block() {
let sql = "BEGIN\n SELECT 1;\n SELECT 2;\nEND";
let cursor = sql[..sql.find("SELECT 2").unwrap()].encode_utf16().count();
assert_eq!(super::find_statement_at_cursor_for_database(sql, cursor, DatabaseType::SqlServer), sql);
}
#[test]
fn sqlserver_current_statement_still_splits_begin_transaction_batches() {
let sql = "BEGIN TRANSACTION;\nUPDATE dbo.T SET x = 1;\nCOMMIT;";
let cursor = sql[..sql.find("UPDATE").unwrap()].encode_utf16().count();
assert_eq!(
super::find_statement_at_cursor_for_database(sql, cursor, DatabaseType::SqlServer),
"UPDATE dbo.T SET x = 1"
);
}
// #13051: the reported script begins with `SET XACT_ABORT ON;`, so the batch
// does not start with the block. The statement at the cursor is still found
// by the enclosing top-level BEGIN TRY ... END CATCH block instead of being
// `;`-split into a lone `BEGIN TRANSACTION`.
#[test]
fn sqlserver_current_statement_keeps_try_catch_after_leading_statement() {
let sql = "\
SET XACT_ABORT ON;
BEGIN TRY
BEGIN TRANSACTION;
UPDATE [dbo].[Documents]
SET [Deleted] = 0
WHERE [Deleted] = 1;
DECLARE @Updated int = @@ROWCOUNT;
IF EXISTS (SELECT 1 FROM [dbo].[Documents] WHERE [Deleted] = 1)
THROW 50001, N'Rows with Deleted = 1 still exist.', 1;
COMMIT TRANSACTION;
SELECT N'OK' AS Result, @Updated AS RowsUpdated;
END TRY
BEGIN CATCH
IF @@TRANCOUNT > 0 ROLLBACK TRANSACTION;
THROW;
END CATCH;";
let block = sql[sql.find("BEGIN TRY").unwrap()..].trim_end_matches(';');
let transaction_cursor = sql[..sql.find("BEGIN TRANSACTION").unwrap()].encode_utf16().count();
let update_cursor = sql[..sql.find("UPDATE [dbo].[Documents]").unwrap()].encode_utf16().count();
let catch_cursor = sql[..sql.find("BEGIN CATCH").unwrap()].encode_utf16().count();
let rollback_cursor = sql[..sql.find("ROLLBACK").unwrap()].encode_utf16().count();
assert_eq!(
super::find_statement_at_cursor_for_database(sql, transaction_cursor, DatabaseType::SqlServer),
block
);
assert_eq!(super::find_statement_at_cursor_for_database(sql, update_cursor, DatabaseType::SqlServer), block);
assert_eq!(super::find_statement_at_cursor_for_database(sql, catch_cursor, DatabaseType::SqlServer), block);
assert_eq!(super::find_statement_at_cursor_for_database(sql, rollback_cursor, DatabaseType::SqlServer), block);
}
// The block boundary must not be capped by a token budget: a block whose
// END is far past the first words still keeps every statement inside it.
#[test]
fn sqlserver_current_statement_keeps_long_top_level_block() {
let body = (0..200).map(|index| format!(" SELECT {index};")).collect::<Vec<_>>().join("\n");
let sql = format!("SET XACT_ABORT ON;\nBEGIN\n{body}\nEND");
let cursor = sql[..sql.find("SELECT 199").unwrap()].encode_utf16().count();
let block = &sql[sql.find("BEGIN").unwrap()..];
assert_eq!(super::find_statement_at_cursor_for_database(&sql, cursor, DatabaseType::SqlServer), block);
let plain = format!("BEGIN\n{body}\nEND");
let cursor = plain[..plain.find("SELECT 199").unwrap()].encode_utf16().count();
assert_eq!(super::find_statement_at_cursor_for_database(&plain, cursor, DatabaseType::SqlServer), plain);
}
#[test]
fn mysql_hash_comments_split_statements_per_issue_428() {
let sql = "SELECT 1; # mysql comment\n\nSELECT 2 # trailing comment";
assert_eq!(
split_sql_statements_for_database(sql, DatabaseType::Mysql),
vec!["SELECT 1", "# mysql comment\n\nSELECT 2 # trailing comment"]
);
}
// #13514: a JDBC SQL Server connection splits with the SQL Server profile
// (the mapped `statement_split_database_type`), so the reported script keeps
// its TRY/CATCH batch whole while the leading `SET` stays its own statement.
#[test]
fn sqlserver_statement_splitter_keeps_try_catch_batch_after_leading_statement() {
let sql = "\
SET XACT_ABORT ON;
BEGIN TRY
BEGIN TRANSACTION;
UPDATE [dbo].[Documents]
SET [Deleted] = 0
WHERE [Deleted] = 1;
DECLARE @Updated int = @@ROWCOUNT;
IF EXISTS (SELECT 1 FROM [dbo].[Documents] WHERE [Deleted] = 1)
THROW 50001, N'Rows with Deleted = 1 still exist.', 1;
COMMIT TRANSACTION;
SELECT N'OK' AS Result, @Updated AS RowsUpdated;
END TRY
BEGIN CATCH
IF @@TRANCOUNT > 0 ROLLBACK TRANSACTION;
THROW;
END CATCH;";
let block = &sql[sql.find("BEGIN TRY").unwrap()..];
assert_eq!(
split_sql_statements_for_database(sql, DatabaseType::SqlServer),
vec!["SET XACT_ABORT ON".to_string(), block.to_string()]
);
}
// The same script through the generic JDBC profile (no SQL Server signals)
// keeps its historical split shape: the Oracle-style routine-body handling
// glues the block opener to the first inner statement, and the rest splits
// per statement. The `statement_split_database_type` mapping is what opts a
// connection into the SQL Server rules instead.
#[test]
fn jdbc_statement_splitter_still_splits_try_catch_batches() {
let sql = "BEGIN TRY\n BEGIN TRANSACTION;\n UPDATE dbo.T SET x = 1;\nEND TRY\nBEGIN CATCH\n THROW;\nEND CATCH;";
assert_eq!(
split_sql_statements_for_database(sql, DatabaseType::Jdbc),
vec![
"BEGIN TRY\n BEGIN TRANSACTION",
"UPDATE dbo.T SET x = 1",
"END TRY\nBEGIN CATCH\n THROW",
"END CATCH",
]
);
}
#[test]
fn sqlserver_statement_splitter_keeps_if_block_batch_whole() {
let sql = "IF NOT EXISTS (SELECT 1 FROM sys.databases WHERE name = N'demo')\nBEGIN\n CREATE DATABASE demo;\nEND;\nSELECT 1;";
assert_eq!(
split_sql_statements_for_database(sql, DatabaseType::SqlServer),
vec![
"IF NOT EXISTS (SELECT 1 FROM sys.databases WHERE name = N'demo')\nBEGIN\n CREATE DATABASE demo;\nEND;",
"SELECT 1",
]
);
}
#[test]
fn sqlserver_statement_splitter_still_splits_begin_transaction_batches() {
let sql = "BEGIN TRANSACTION;\nUPDATE dbo.T SET x = 1;\nCOMMIT TRANSACTION;";
assert_eq!(
split_sql_statements_for_database(sql, DatabaseType::SqlServer),
vec!["BEGIN TRANSACTION", "UPDATE dbo.T SET x = 1", "COMMIT TRANSACTION"]
);
}
// The four specs below mirror `sqlStatementRanges.spec.ts` so the backend
// splitter and the editor range scanner agree on where a control-flow
// batch ends when its final `END` has no semicolon of its own.
#[test]
fn sqlserver_statement_splitter_ends_single_line_if_batch_before_next_statement() {
let sql = "IF @x = 1 BEGIN SELECT 1; END\nSELECT 999;";
assert_eq!(
split_sql_statements_for_database(sql, DatabaseType::SqlServer),
vec!["IF @x = 1 BEGIN SELECT 1; END", "SELECT 999"]
);
}
#[test]
fn sqlserver_statement_splitter_ends_try_batch_without_catch_at_end_try() {
let sql = "BEGIN TRY\n SELECT 1;\nEND TRY\nSELECT 999;";
assert_eq!(
split_sql_statements_for_database(sql, DatabaseType::SqlServer),
vec!["BEGIN TRY\n SELECT 1;\nEND TRY", "SELECT 999"]
);
}
#[test]
fn sqlserver_statement_splitter_ends_if_else_batch_before_next_statement() {
let batch = "\
IF NOT EXISTS (SELECT 1 FROM dbo.T WHERE n = N'x')
BEGIN
SELECT 1;
END
ELSE
BEGIN
SELECT 2;
END";
assert_eq!(
split_sql_statements_for_database(&format!("{batch}\nSELECT 999;"), DatabaseType::SqlServer),
vec![batch.to_string(), "SELECT 999".to_string()]
);
}
#[test]
fn sqlserver_statement_splitter_ends_if_else_batch_at_else_branch_end_without_semicolon() {
let batch = "\
IF @x = 1
BEGIN
SELECT 1;
END
ELSE
BEGIN
SELECT 2
END";
assert_eq!(
split_sql_statements_for_database(&format!("{batch}\nSELECT 999;"), DatabaseType::SqlServer),
vec![batch.to_string(), "SELECT 999".to_string()]
);
}
#[test]
fn sqlserver_statement_splitter_keeps_case_end_inside_if_batch() {
// `CASE ... END` must not close the surrounding `BEGIN` block, so the
// batch still ends at its own `END`, before the next statement.
let sql = "IF @x = 1 BEGIN SELECT CASE WHEN 1 = 1 THEN 'a' ELSE 'b' END AS v; END\nSELECT 999;";
assert_eq!(
split_sql_statements_for_database(sql, DatabaseType::SqlServer),
vec!["IF @x = 1 BEGIN SELECT CASE WHEN 1 = 1 THEN 'a' ELSE 'b' END AS v; END", "SELECT 999",]
);
}
#[test]
fn sqlserver_statement_splitter_ignores_end_conversation_in_batches() {
let sql = "IF @x = 1 BEGIN SELECT 1; END\nEND CONVERSATION @handle;\nSELECT 999;";
assert_eq!(
split_sql_statements_for_database(sql, DatabaseType::SqlServer),
vec!["IF @x = 1 BEGIN SELECT 1; END", "END CONVERSATION @handle", "SELECT 999",]
);
}
#[test]
fn sqlserver_statement_splitter_keeps_nested_try_catch_batch_whole() {
let batch = "\
WHILE @i < 10
BEGIN
BEGIN TRY
SELECT 1;
END TRY
BEGIN CATCH
SET @i = @i + 1;
END CATCH
END";
let sql = format!("{batch}\nSELECT 999;");
assert_eq!(
split_sql_statements_for_database(&sql, DatabaseType::SqlServer),
vec![batch.to_string(), "SELECT 999".to_string()]
);
}
#[test]
fn mysql_delimiter_command_keeps_procedure_body_together_per_issue_1978() {
let sql = "\
-- ----------------------------
-- Procedure structure for fix_collation
-- ----------------------------
DROP PROCEDURE IF EXISTS `fix_collation`;
delimiter ;;
CREATE PROCEDURE `fix_collation`()
BEGIN
DECLARE done INT DEFAULT FALSE;
DECLARE tbl_name VARCHAR(255);
DECLARE cur CURSOR FOR
SELECT TABLE_NAME FROM information_schema.TABLES
WHERE TABLE_SCHEMA = DATABASE() AND TABLE_COLLATION = 'utf8mb4_0900_ai_ci';
DECLARE CONTINUE HANDLER FOR NOT FOUND SET done = TRUE;
OPEN cur;
read_loop: LOOP
FETCH cur INTO tbl_name;
IF done THEN LEAVE read_loop; END IF;
SET @sql = CONCAT('ALTER TABLE `', tbl_name, '` CONVERT TO CHARACTER SET utf8mb4 COLLATE utf8mb4_general_ci');
PREPARE stmt FROM @sql;
EXECUTE stmt;
DEALLOCATE PREPARE stmt;
END LOOP;
CLOSE cur;
END
;;
delimiter ;";
assert_eq!(
split_sql_statements_for_database(sql, DatabaseType::Mysql),
vec![
"-- ----------------------------\n-- Procedure structure for fix_collation\n-- ----------------------------\nDROP PROCEDURE IF EXISTS `fix_collation`",
"CREATE PROCEDURE `fix_collation`()\nBEGIN\nDECLARE done INT DEFAULT FALSE;\nDECLARE tbl_name VARCHAR(255);\nDECLARE cur CURSOR FOR\nSELECT TABLE_NAME FROM information_schema.TABLES\nWHERE TABLE_SCHEMA = DATABASE() AND TABLE_COLLATION = 'utf8mb4_0900_ai_ci';\nDECLARE CONTINUE HANDLER FOR NOT FOUND SET done = TRUE;\n\n OPEN cur;\n read_loop: LOOP\n FETCH cur INTO tbl_name;\n IF done THEN LEAVE read_loop; END IF;\n SET @sql = CONCAT('ALTER TABLE `', tbl_name, '` CONVERT TO CHARACTER SET utf8mb4 COLLATE utf8mb4_general_ci');\n PREPARE stmt FROM @sql;\n EXECUTE stmt;\n DEALLOCATE PREPARE stmt;\n END LOOP;\n CLOSE cur;\nEND",
]
);
}
#[test]
fn mysql_current_statement_ignores_delimiter_command_semicolon_per_issue_1978() {
let sql = "\
delimiter ;;
CREATE PROCEDURE `fix_collation`()
BEGIN
SET @sql = CONCAT('ALTER TABLE `', 't', '` CONVERT TO CHARACTER SET utf8mb4 COLLATE utf8mb4_general_ci');
PREPARE stmt FROM @sql;
EXECUTE stmt;
END
;;
delimiter ;";
let cursor = sql[..sql.find("PREPARE").unwrap()].encode_utf16().count();
assert_eq!(
find_statement_at_cursor_for_database(sql, cursor, DatabaseType::Mysql),
"CREATE PROCEDURE `fix_collation`()\nBEGIN\n SET @sql = CONCAT('ALTER TABLE `', 't', '` CONVERT TO CHARACTER SET utf8mb4 COLLATE utf8mb4_general_ci');\n PREPARE stmt FROM @sql;\n EXECUTE stmt;\nEND"
);
}
#[test]
fn mysql_repeated_delimiter_command_does_not_emit_directive_or_panic() {
let sql = r#"DELIMITER ;;
drop procedure if exists p_ludp_data_table_scripts;;
delimiter ;;
create procedure p_ludp_data_table_scripts()
begin
-- 定义变量,存储表及字段存在状态
declare filedExist INT DEFAULT 0;
declare currentDatabaseName VARCHAR(200) DEFAULT '';
declare tableCount INT DEFAULT 0;
end ;;
call p_ludp_data_table_scripts();"#;
let expected = vec![
"drop procedure if exists p_ludp_data_table_scripts",
"create procedure p_ludp_data_table_scripts()\nbegin\n -- 定义变量,存储表及字段存在状态\n declare filedExist INT DEFAULT 0;\n declare currentDatabaseName VARCHAR(200) DEFAULT '';\n declare tableCount INT DEFAULT 0;\n\nend",
"call p_ludp_data_table_scripts();",
];
assert_eq!(split_sql_statements_for_database(sql, DatabaseType::Mysql), expected);
let ranges =
split_sql_statement_ranges_with_options(sql, SqlParsingOptions::for_database_type(DatabaseType::Mysql));
assert_eq!(ranges.iter().map(|range| range.text.as_str()).collect::<Vec<_>>(), expected);
let cursor = sql[..sql.find("declare currentDatabaseName").unwrap()].encode_utf16().count();
assert_eq!(find_statement_at_cursor_for_database(sql, cursor, DatabaseType::Mysql), expected[1]);
let mut splitter =
SqlStatementSplitter::with_options(SqlParsingOptions::for_database_type(DatabaseType::Mysql));
let mut streamed = Vec::new();
for ch in sql.chars() {
streamed.extend(splitter.push_chunk(&ch.to_string()));
}
streamed.extend(splitter.finish());
assert_eq!(streamed, expected);
}
#[test]
fn mysql_delimiter_command_skips_empty_custom_delimiter_statement_per_issue_1988() {
let sql = "\
select COUNT(1) FROM your_table;
delimiter ;;
select COUNT(1) FROM your_table;
;;
delimiter ;";
assert_eq!(
split_sql_statements_for_database(sql, DatabaseType::Mysql),
vec!["select COUNT(1) FROM your_table", "select COUNT(1) FROM your_table;"]
);
}
#[test]
fn mysql_delimiter_command_ranges_skip_empty_custom_delimiter_statement_per_issue_1988() {
let sql = "\
select COUNT(1) FROM your_table;
delimiter ;;
select COUNT(1) FROM your_table;
;;
delimiter ;";
let ranges =
split_sql_statement_ranges_with_options(sql, SqlParsingOptions::for_database_type(DatabaseType::Mysql));
assert_eq!(
ranges.iter().map(|range| range.text.as_str()).collect::<Vec<_>>(),
vec!["select COUNT(1) FROM your_table", "select COUNT(1) FROM your_table;"]
);
}
#[test]
fn mysql_current_statement_keeps_inline_hash_comment_per_issue_428() {
let sql = "SELECT 1; # mysql comment\n\nSELECT 2 # trailing comment";
let cursor = sql[..sql.find("SELECT 2").unwrap()].encode_utf16().count();
assert_eq!(
find_statement_at_cursor_for_database(sql, cursor, DatabaseType::Mysql),
"SELECT 2 # trailing comment"
);
}
#[test]
fn mysql_current_statement_preserves_only_exact_tdsql_proxy_directive() {
let sql = "SELECT 1;\n/*proxy*/ \n\tSHOW PROXY STATUS;\nSELECT 2;";
let cursor = sql[..sql.find("STATUS").unwrap()].encode_utf16().count();
assert_eq!(
find_statement_at_cursor_for_database(sql, cursor, DatabaseType::Mysql),
"/*proxy*/ \n\tSHOW PROXY STATUS"
);
assert_eq!(
split_sql_statements_for_database(sql, DatabaseType::Mysql),
vec!["SELECT 1", "/*proxy*/ \n\tSHOW PROXY STATUS", "SELECT 2"]
);
for inline in [
"/*proxy*/SHOW PROXY STATUS",
"/* ordinary */\n/*proxy*/SHOW PROXY STATUS",
"/*+ route */\n/*proxy*/SHOW PROXY STATUS",
] {
let cursor = inline[..inline.find("STATUS").unwrap()].encode_utf16().count();
assert_eq!(
find_statement_at_cursor_for_database(inline, cursor, DatabaseType::Mysql),
"/*proxy*/SHOW PROXY STATUS"
);
}
assert!(split_sql_statements_for_database("/*proxy*/", DatabaseType::Mysql).is_empty());
}
#[test]
fn mysql_current_statement_preserves_same_line_tdsql_directives_without_an_allowlist() {
for directive in ["/*sets:allsets */", "/*master*/", "/*slave:set_1781591902_7*/", "/*future-route:anywhere*/"]
{
let sql = format!("{directive} SELECT count(*) FROM tenant_table");
let cursor = sql[..sql.find("tenant_table").unwrap()].encode_utf16().count();
assert_eq!(find_statement_at_cursor_for_database(&sql, cursor, DatabaseType::Mysql), sql);
assert_eq!(split_sql_statements_for_database(&sql, DatabaseType::Mysql), vec![sql]);
}
}
#[test]
fn mysql_current_statement_strips_generic_tdsql_style_directive_on_a_separate_line() {
let sql = "/*sets:allsets */\nSELECT count(*) FROM tenant_table";
let cursor = sql[..sql.find("tenant_table").unwrap()].encode_utf16().count();
assert_eq!(
find_statement_at_cursor_for_database(sql, cursor, DatabaseType::Mysql),
"SELECT count(*) FROM tenant_table"
);
}
#[test]
fn other_databases_strip_same_line_tdsql_style_directives() {
let sql = "/*sets:allsets */ SELECT count(*) FROM tenant_table";
let cursor = sql[..sql.find("tenant_table").unwrap()].encode_utf16().count();
assert_eq!(
find_statement_at_cursor_for_database(sql, cursor, DatabaseType::Postgres),
"SELECT count(*) FROM tenant_table"
);
}
#[test]
fn db2_current_statement_skips_complete_nested_comments() {
let sql = "/* outer /* inner */ trailing */ SELECT IBMREQD FROM SYSIBM.SYSDUMMY1;";
let cursor = sql.encode_utf16().count();
assert_eq!(
find_statement_at_cursor_for_database(sql, cursor, DatabaseType::Db2),
"SELECT IBMREQD FROM SYSIBM.SYSDUMMY1"
);
let script = "/* outer /* SELECT 99; */ DELETE FROM hidden; */ VALUES 1; VALUES 2;";
assert_eq!(
split_sql_statements_for_database(script, DatabaseType::Db2),
vec!["/* outer /* SELECT 99; */ DELETE FROM hidden; */ VALUES 1", "VALUES 2"]
);
}
#[test]
fn mysql_current_statement_strips_other_leading_block_comments() {
for comment in ["/* ordinary */", "/*unknown*/", "/* proxy */", "/*PROXY*/"] {
let sql = format!("{comment}\nSHOW PROXY STATUS");
let cursor = sql[..sql.find("STATUS").unwrap()].encode_utf16().count();
assert_eq!(find_statement_at_cursor_for_database(&sql, cursor, DatabaseType::Mysql), "SHOW PROXY STATUS");
}
for separated in ["/*proxy*/\n/* audit */\nSHOW PROXY STATUS", "/*proxy*/\n-- audit\nSHOW PROXY STATUS"] {
let cursor = separated[..separated.find("STATUS").unwrap()].encode_utf16().count();
assert_eq!(
find_statement_at_cursor_for_database(separated, cursor, DatabaseType::Mysql),
"SHOW PROXY STATUS"
);
}
let postgres = "/*proxy*/\nSHOW PROXY STATUS";
let cursor = postgres[..postgres.find("STATUS").unwrap()].encode_utf16().count();
assert_eq!(
find_statement_at_cursor_for_database(postgres, cursor, DatabaseType::Postgres),
"SHOW PROXY STATUS"
);
let followed_by_hint = "/*proxy*/\n/*+ route */\nSHOW PROXY STATUS";
let cursor = followed_by_hint[..followed_by_hint.find("STATUS").unwrap()].encode_utf16().count();
assert_eq!(
find_statement_at_cursor_for_database(followed_by_hint, cursor, DatabaseType::Mysql),
"SHOW PROXY STATUS"
);
}
#[test]
fn mysql_single_statement_with_inline_comment_stays_executable_per_issue_428() {
let sql = "SELECT 1 # mysql comment";
let cursor = sql.encode_utf16().count();
assert_eq!(find_statement_at_cursor_for_database(sql, cursor, DatabaseType::Mysql), "SELECT 1 # mysql comment");
}
}
#[cfg(test)]
mod editor_encoding_tests {
use super::*;
#[test]
fn utf16_bytes_and_round_trip() {
for (encoding, expected) in [
("utf16le", vec![0xff, 0xfe, 0x41, 0, 0x2d, 0x4e, 0x3d, 0xd8, 0, 0xde]),
("utf16be", vec![0xfe, 0xff, 0, 0x41, 0x4e, 0x2d, 0xd8, 0x3d, 0xde, 0]),
] {
let bytes = encode_sql_file_text("A中😀", encoding).unwrap();
assert_eq!(bytes, expected);
assert_eq!(decode_sql_file_bytes_with_encoding(&bytes, encoding).unwrap(), "A中😀");
}
}
}