Apache DataFusion SQL Query Engine
perf: optimize strpos by eliminating double iteration for UTF-8 (#19572)
Liang-Chi Hsieh195d3d64bb9f
diff --git a/datafusion/functions/src/unicode/strpos.rs b/datafusion/functions/src/unicode/strpos.rsindex 1f980de9e8..9be086c4cf 100644--- a/datafusion/functions/src/unicode/strpos.rs+++ b/datafusion/functions/src/unicode/strpos.rs@@ -215,14 +215,37 @@ where)}} else {- // The `find` method returns the byte index of the substring.- // We count the number of chars up to that byte index.- T::Native::from_usize(- string- .find(substring)- .map(|x| string[..x].chars().count() + 1)- .unwrap_or(0),- )+ // For non-ASCII, use a single-pass search that tracks both+ // byte position and character position simultaneously+ if substring.is_empty() {+ return T::Native::from_usize(1);+ }++ let substring_bytes = substring.as_bytes();+ let string_bytes = string.as_bytes();++ if substring_bytes.len() > string_bytes.len() {+ return T::Native::from_usize(0);+ }++ // Single pass: find substring while counting characters+ let mut char_pos = 0;+ for (byte_idx, _) in string.char_indices() {+ char_pos += 1;+ if byte_idx + substring_bytes.len() <= string_bytes.len() {+ // SAFETY: We just checked that byte_idx + substring_bytes.len() <= string_bytes.len()+ let slice = unsafe {+ string_bytes.get_unchecked(+ byte_idx..byte_idx + substring_bytes.len(),+ )+ };+ if slice == substring_bytes {+ return T::Native::from_usize(char_pos);+ }+ }+ }++ T::Native::from_usize(0)}}_ => None,