Skip to content

armgit / cindy

Public Imported from GitHub
0

Consider it done. The open-source AI agent that works out of the box · 想到,就能做到。开源、开箱即用的 AI Agent。

fix(shared): scan source context without regex backtracking

DavidShen66ade59401ff
diff --git a/packages/maker-shared/src/__tests__/pluginResults.test.ts b/packages/maker-shared/src/__tests__/pluginResults.test.ts
index 5ff542d1b5..0e18af7b12 100644
--- a/packages/maker-shared/src/__tests__/pluginResults.test.ts
+++ b/packages/maker-shared/src/__tests__/pluginResults.test.ts
@@ -4,6 +4,17 @@ import { hasVisibleHistoryResult } from '../historyViewProjection.js';
const blob = (ext: string) => `cindy-media://blobs/${'a'.repeat(64)}.${ext}`;
describe('portable plugin results', () => {
+ it.each(["'", '"', String.fromCharCode(96)])('handles long escaped source literals ending in a backslash: %s', (quote) => {
+ const source = 'const message = ' + quote + ('\\' + quote).repeat(5000)
+ + ' Open xdt-file:///tmp/fixture.pdf' + '\\';
+ expect(files(source)).toEqual([]);
+ expect(files(JSON.stringify({ text: source }))).toEqual([]);
+ });
+ it('handles repeated unterminated block-comment openers without hiding earlier real files', () => {
+ const url = 'xdt-file:///tmp/report.pdf';
+ const source = 'Saved [report](' + url + ')\n/*' + 'a/*'.repeat(5000);
+ expect(files(source)).toEqual([{ url, title: 'report.pdf' }]);
+ });
it.each(["'", '"', String.fromCharCode(96)])('ignores URLs anywhere inside source literals quoted with %s', (quote) => {
const url = 'xdt-file:///tmp/fixture.pdf';
const body = 'Open ' + url + ' or [report](xdt-file://open?path=%2Ftmp%2Fother.pdf)';
diff --git a/packages/maker-shared/src/payloadSummary.ts b/packages/maker-shared/src/payloadSummary.ts
index 5e920abd95..b1164042f1 100644
--- a/packages/maker-shared/src/payloadSummary.ts
+++ b/packages/maker-shared/src/payloadSummary.ts
@@ -488,15 +488,53 @@ function absoluteToolFilePath(url: string): string | null {
}
}
-const SOURCE_CONTEXT_TOKEN_RE = /"(?:\\[\s\S]|[^"\\])*(?:"|$)|'(?:\\[\s\S]|[^'\\])*(?:'|$)|`(?:\\[\s\S]|[^`\\])*(?:`|$)|\/\*[\s\S]*?\*\/|(?<![:/])\/\/[^\r\n]*/g;
+function* sourceContextTokens(text: string): Generator<{
+ start: number; end: number; comment: boolean; closed: boolean;
+}> {
+ // Always advance past a token, including unterminated strings/comments.
+ // An unanchored regex can retry every escaped quote or /* in a long result.
+ let index = 0;
+ while (index < text.length) {
+ const start = index;
+ const quote = text[index];
+ if (quote === '"' || quote === "'" || quote === '`') {
+ index++;
+ let closed = false;
+ while (index < text.length) {
+ if (text[index] === '\\') {
+ index = Math.min(index + 2, text.length);
+ } else if (text[index++] === quote) {
+ closed = true;
+ break;
+ }
+ }
+ yield { start, end: index, comment: false, closed };
+ } else if (text.startsWith('/*', index)) {
+ const closing = text.indexOf('*/', index + 2);
+ index = closing < 0 ? text.length : closing + 2;
+ yield { start, end: index, comment: true, closed: closing >= 0 };
+ } else if (text.startsWith('//', index) && text[index - 1] !== ':' && text[index - 1] !== '/') {
+ index += 2;
+ while (index < text.length && text[index] !== '\r' && text[index] !== '\n') index++;
+ yield { start, end: index, comment: true, closed: true };
+ } else {
+ index++;
+ }
+ }
+}
function withoutSourceComments(text: string): string {
// Skip string literals before recognizing comments, so URL schemes
// and comment-like file names inside strings remain part of the context.
- return text.replace(
- SOURCE_CONTEXT_TOKEN_RE,
- (token) => token.startsWith('/') ? token.replace(/[^\r\n]/g, ' ') : token,
- );
+ const parts: string[] = [];
+ let end = 0;
+ for (const token of sourceContextTokens(text)) {
+ if (!token.comment) continue;
+ parts.push(text.slice(end, token.start), text.slice(token.start, token.end).replace(/[^\r\n]/g, ' '));
+ end = token.end;
+ }
+ parts.push(text.slice(end));
+ return parts.join('');
}
function looksLikeQuotedSourceLiteral(text: string, index: number, end: number): boolean {
@@ -518,12 +556,11 @@ function* quotedSourceRanges(text: string): Generator<{ start: number; end: numb
// Classify the whole literal, not only URLs touching its opening quote.
// Comments and escaped quotes use the same token boundaries as source context.
let previous: { end: number; source: boolean } | undefined;
- for (const match of text.matchAll(SOURCE_CONTEXT_TOKEN_RE)) {
- if (match[0].startsWith('/')) continue;
- const start = match.index;
+ for (const token of sourceContextTokens(text)) {
+ if (token.comment) continue;
+ const start = token.start;
// Tool output may end mid-literal after the Host applies its byte budget.
- const closed = match[0].length > 1 && match[0].endsWith(match[0][0]);
- const end = start + match[0].length - (closed ? 1 : 0);
+ const end = token.end - (token.closed ? 1 : 0);
const continuesList = previous
&& withoutSourceComments(text.slice(previous.end + 1, start)).trim() === ',';
const source: boolean = previous && continuesList