From 57d38d69cba09f17d0f979fe89fb36f9cdbf6be3 Mon Sep 17 00:00:00 2001 From: seekskyworld Date: Thu, 17 Sep 2026 23:47:17 +0800 Subject: [PATCH] perf(runtime): bound Read line-index memory to the page Count total lines without retaining the full file index, and keep only positions inside the bounded response window. Preserve exact range and continuation semantics. Fixes #5460 Generated-by: OpenAI Codex Signed-off-by: seekskyworld --- .../runtime/src/__tests__/read-page.test.ts | 46 +++++++++++++++++++ packages/runtime/src/read-page.ts | 40 ++++++++++++---- 2 files changed, 76 insertions(+), 10 deletions(-) diff --git a/packages/runtime/src/__tests__/read-page.test.ts b/packages/runtime/src/__tests__/read-page.test.ts index 3137fb48a3..cc7f4f86d1 100644 --- a/packages/runtime/src/__tests__/read-page.test.ts +++ b/packages/runtime/src/__tests__/read-page.test.ts @@ -18,6 +18,8 @@ */ import assert from 'node:assert/strict'; +import { execFile } from 'node:child_process'; +import { promisify } from 'node:util'; import { test } from 'node:test'; import { readPage, @@ -175,3 +177,47 @@ test('archive Read preserves fields beside structured text across pages', () => assert.deepEqual(JSON.parse(recovered), value); } }); + +test('small line ranges do not need a whole-file line index', async () => { + // 固定堆预算验证实际资源边界，不依赖耗时阈值或具体索引实现。 + const { stdout } = await promisify(execFile)( + process.execPath, + [ + '--max-old-space-size=64', + '--input-type=module', + '--eval', + ` + import assert from 'node:assert/strict'; + import { readPage } from ${JSON.stringify(new URL('../read-page.js', import.meta.url).href)}; + const content = 'x\\n'.repeat(8_000_000); + const page = readPage(content, { path: 'log.txt', offset: 2_000_000, limit: 20 }); + assert.equal(page.totalLines, 8_000_001); + assert.equal(page.offset, 2_000_000); + assert.equal(page.returnedLines, 20); + assert.equal(page.content, 'x\\n'.repeat(19) + 'x'); + assert.equal(page.next, null); + console.log('bounded read passed'); + `, + ], + { timeout: 30_000 }, + ); + assert.equal(stdout.trim(), 'bounded read passed'); +}); + +test('line ranges preserve empty lines, trailing newlines and offsets beyond EOF', () => { + for (const content of ['', '\n', '\n\n', 'a\nb', 'a\nb\n', 'a\r\n😀\n\nlast']) { + const lines = content.split('\n'); + for (let offset = 0; offset <= lines.length + 1; offset++) { + for (const limit of [undefined, 1, 2, 100]) { + const selected = lines.slice(offset, limit === undefined ? undefined : offset + limit); + assert.deepEqual(readPage(content, { path: 'file.txt', offset, limit }), { + content: selected.join('\n'), + offset, + returnedLines: selected.length, + totalLines: lines.length, + next: null, + }); + } + } + } +}); diff --git a/packages/runtime/src/read-page.ts b/packages/runtime/src/read-page.ts index d1361f9934..aa7e231d54 100644 --- a/packages/runtime/src/read-page.ts +++ b/packages/runtime/src/read-page.ts @@ -119,10 +119,34 @@ export function readPage( 'Invalid Read continuation position. Copy the complete next object from the preceding result.', ); } - const starts = [0]; - for (let at = content.indexOf('\n'); at >= 0; at = content.indexOf('\n', at + 1)) + // 总行数仍需扫描全文，但只为本页可能返回的字符保留行索引，避免短行日志 + // 为一个小响应分配与整个文件行数成正比的临时数组。 + let totalLines = 1; + let offset = resolved.position === undefined ? (input.offset ?? 0) : 0; + let start = resolved.position ?? (offset === 0 ? 0 : content.length); + let lineStart = 0; + let end = content.length; + for (let at = content.indexOf('\n'); at >= 0; at = content.indexOf('\n', at + 1)) { + const line = totalLines++; + if (resolved.position !== undefined && at < resolved.position) { + offset = line; + lineStart = at + 1; + } else if (resolved.position === undefined && line === offset) { + start = at + 1; + lineStart = start; + } + if (input.limit !== undefined && line === offset + input.limit && at >= start) end = at; + } + const lastLine = Math.min(totalLines, offset + (input.limit ?? totalLines)); + const starts = [lineStart]; + const indexEnd = Math.min(end, start + maxChars); + for ( + let at = content.indexOf('\n', start); + at >= 0 && at < indexEnd; + at = content.indexOf('\n', at + 1) + ) { starts.push(at + 1); - const totalLines = starts.length; + } const lineAt = (position: number): number => { let low = 0; let high = starts.length - 1; @@ -131,17 +155,13 @@ export function readPage( if (starts[middle]! <= position) low = middle; else high = middle - 1; } - return low; + return offset + low; }; - const offset = resolved.position === undefined ? (input.offset ?? 0) : lineAt(resolved.position); - const start = resolved.position ?? starts[Math.min(offset, totalLines)] ?? content.length; - const lastLine = Math.min(totalLines, offset + (input.limit ?? totalLines)); - const end = lastLine < totalLines ? starts[lastLine]! - 1 : content.length; let digest = resolved.digest; const makePage = (stop: number): ReadPage => { const complete = stop >= end; const nextLine = Math.max(offset, lineAt(stop)); - const boundary = stop === (starts[nextLine + 1] ?? content.length + 1) - 1; + const boundary = content[stop] === '\n'; const partialLine = !complete && !boundary; const returnedLines = offset >= totalLines @@ -163,7 +183,7 @@ export function readPage( offset, returnedLines, totalLines, - ...(partialLine || (offset < totalLines && start > starts[offset]!) + ...(partialLine || (offset < totalLines && start > lineStart) ? { partialLine: true as const } : {}), next,