fix: avoid quadratic re-scan of comments after a value#1689
Merged
baylesj merged 2 commits intoJun 14, 2026
Conversation
OurReader::readComment() decides whether a comment should be attached to the previous value (commentAfterOnSameLine) by scanning the input from the end of that value up to the comment with containsNewLine(). lastValueEnd_ only advances when a new value is read, so a long run of comments after a value (e.g. during error recovery, or a value followed by many comments) made every comment re-scan the same growing prefix, giving O(n^2) parse time. A jsoncpp_fuzzer testcase took ~18s for a 400KB input. A comment can only ever be on the same line as the last value if no newline separates them, and the gap to inspect only grows as further comments are consumed, so once the gap has been examined for the first comment it never needs to be examined again. Mark lastValueHasAComment_ after the first comment following a value so subsequent comments skip the scan. Parsing the testcase drops from ~18s to ~56ms with identical output. Add a regression test that parses a value followed by a large number of trailing comments and requires it to complete well under a generous time bound.
Replace the wall-clock bound in the comment regression test with a direct, deterministic assertion on work done. The parse output is identical with and without the fix, so the only observable difference is how much the parser scans; a time bound is also flaky under valgrind/sanitizers/loaded CI. Add an instrumentation counter for the bytes examined by OurReader::containsNewLine, exposed via a JSON_API seam, and assert it stays linear in the input (scanned < 4 * doc.size()) rather than O(comments * gap). The counter is thread_local (no race during concurrent parsing) and the increment is negligible, running only while parsing comments. It is compiled unconditionally because the ABI compatibility job builds the test suite against a separately-installed Release library, so the symbol must exist there. Rename the test to parseCommentsAfterValueScansLinearly to describe what it checks, and link crbug.com/521541633. Verified: the test fails when the fix is reverted and passes with it, in Debug and Release, and the seam links against a Release-installed shared library (the ABI compatibility scenario).
Contributor
|
The production changes are good, but I'm reluctant to land a non-deterministic timer based test. Updating the test format to be based on execution count instead of a potentially flaky clock timer, and then landing this change. Thank you for your CL! |
baylesj
force-pushed
the
fix-quadratic-comment-rescan
branch
from
1eeb764 to
d0c03d1
Compare
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
OurReader::readComment()decides whether a comment attaches to the previous value (commentAfterOnSameLine) by scanning from the end of that value to the comment viacontainsNewLine(lastValueEnd_, commentBegin).lastValueEnd_only advances when a new value is read, so a long run of comments after a value (error recovery, or a value followed by many comments) re-scans the same growing prefix for every comment -- O(n^2).A comment can only be on the same line as the last value when no newline separates them, and that gap only grows as more comments are consumed, so it never needs re-examining after the first comment. Setting
lastValueHasAComment_after the first comment makes the rest skip the scan. Output is unchanged.Found via a
jsoncpp_fuzzertimeout (Chromium issue 521541633): a 400KB input scanned 2.24 GB across 8384containsNewLinecalls and took ~18s. After the fix: 122 KB / 1 call, ~56ms.Adds
CharReaderTest/parseCommentsAfterValueNotQuadratic(a value followed by many trailing comments, bounded parse time): fails multi-second without the fix, passes in ms with it. Existing tests unaffected.