perf(json): make deeply nested JSON.stringify O(depth^2)#887
Merged
Conversation
MakeIndent built the pretty-print indent with `Result := Result + FGap` in a loop (O(level^2) per call) and was called twice per object/array node, so deeply nested values spent O(depth^3) building indentation. Replace the accumulator loop with DupeString(FGap, ALevel), a single SetLength+Move pass (O(level)). Output is byte-identical (ES2026 §25.5.4.5/§25.5.4.6: the indent at depth d is the gap repeated d times), and DupeString returns '' for level <= 0 and for an empty gap, so the previous guards are no longer needed. Add exact-output regression tests (deep object, deep array, multi-char gap, depth-30 per-level indent width) and depth-50 indent benchmarks for both object and array nesting. Closes #811 Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
|
The latest updates on your projects. Learn more about Vercel for GitHub. 1 Skipped Deployment
|
Contributor
|
No actionable comments were generated in the recent review. 🎉 ℹ️ Recent review info⚙️ Run configurationConfiguration used: Organization UI Review profile: CHILL Plan: Pro Plus Run ID: 📒 Files selected for processing (4)
📝 WalkthroughWalkthroughRefactors JSON Stringify Buffered Writer Refactor
Estimated code review effort🎯 3 (Moderate) | ⏱️ ~25 minutes Possibly related PRs
🚥 Pre-merge checks | ✅ 5✅ Passed checks (5 passed)
Comment |
Contributor
Suite TimingTest Runner (interpreted: 10,956 passed; bytecode: 10,956 passed)
MemoryGC rows aggregate the main thread plus all worker thread-local GCs. Test runner worker shutdown frees thread-local heaps in bulk; that shutdown reclamation is not counted as GC collections or collected objects.
Benchmarks (interpreted: 433; bytecode: 433)
MemoryGC rows aggregate the main thread plus all worker thread-local GCs. Benchmark runner performs explicit between-file collections, so collection and collected-object counts can be much higher than the test runner.
Measured on ubuntu-latest x64. |
Contributor
Benchmark Results433 benchmarks · PR vs same-runner Interpreted: 🟢 23 improved · 🔴 35 regressed · 375 unchanged · avg +1.3% Typical per-run noise (median variance): interpreted ±3.2%, bytecode ±2.7%. Deltas within noise overlap and read as unchanged. arraybuffer.js — Interp: 🔴 3, 11 unch. · avg -2.6% · Bytecode: 🔴 3, 11 unch. · avg -1.5%
arrays.js — Interp: 19 unch. · avg -1.2% · Bytecode: 🟢 1, 🔴 2, 16 unch. · avg -1.9%
async-await.js — Interp: 6 unch. · avg +0.9% · Bytecode: 6 unch. · avg -2.7%
async-generators.js — Interp: 2 unch. · avg +1.9% · Bytecode: 2 unch. · avg -4.1%
atomics.js — Interp: 🔴 1, 5 unch. · avg +0.2% · Bytecode: 🟢 1, 5 unch. · avg +3.9%
base64.js — Interp: 🔴 1, 9 unch. · avg -1.4% · Bytecode: 10 unch. · avg -0.6%
classes.js — Interp: 🟢 2, 🔴 2, 27 unch. · avg -0.1% · Bytecode: 🟢 2, 🔴 4, 25 unch. · avg -0.5%
closures.js — Interp: 11 unch. · avg +0.8% · Bytecode: 11 unch. · avg +0.1%
collections.js — Interp: 🔴 2, 10 unch. · avg +1.3% · Bytecode: 🟢 1, 11 unch. · avg +3.0%
csv.js — Interp: 13 unch. · avg -2.0% · Bytecode: 13 unch. · avg +1.6%
destructuring.js — Interp: 🟢 3, 19 unch. · avg +0.9% · Bytecode: 🟢 3, 🔴 1, 18 unch. · avg +1.9%
fibonacci.js — Interp: 8 unch. · avg +1.7% · Bytecode: 🔴 2, 6 unch. · avg -3.9%
float16array.js — Interp: 🟢 2, 🔴 1, 29 unch. · avg +0.3% · Bytecode: 🟢 6, 26 unch. · avg +2.0%
for-of.js — Interp: 🔴 1, 6 unch. · avg -1.9% · Bytecode: 7 unch. · avg -2.0%
generators.js — Interp: 🔴 1, 3 unch. · avg +1.0% · Bytecode: 4 unch. · avg +1.1%
intl.js — Interp: 🟢 2, 4 unch. · avg +5.9% · Bytecode: 🔴 1, 5 unch. · avg -4.1%
iterators.js — Interp: 🟢 3, 🔴 1, 38 unch. · avg +2.5% · Bytecode: 🟢 3, 🔴 3, 36 unch. · avg -0.1%
json.js — Interp: 🟢 4, 19 unch. · avg +32.6% · Bytecode: 🟢 4, 🔴 2, 17 unch. · avg +26.7%
jsx.jsx — Interp: 🔴 2, 19 unch. · avg -2.1% · Bytecode: 🟢 4, 🔴 4, 13 unch. · avg +1.7%
modules.js — Interp: 🔴 2, 7 unch. · avg -5.3% · Bytecode: 🟢 1, 8 unch. · avg +1.7%
numbers.js — Interp: 11 unch. · avg +0.1% · Bytecode: 🟢 2, 🔴 1, 8 unch. · avg +2.2%
objects.js — Interp: 🔴 1, 6 unch. · avg -2.7% · Bytecode: 🔴 1, 6 unch. · avg -1.3%
promises.js — Interp: 12 unch. · avg -2.2% · Bytecode: 🔴 2, 10 unch. · avg -0.3%
property-access.js — Interp: 🔴 1, 4 unch. · avg -5.6% · Bytecode: 🔴 1, 4 unch. · avg -2.9%
regexp.js — Interp: 11 unch. · avg +0.7% · Bytecode: 🟢 1, 10 unch. · avg +1.3%
strings.js — Interp: 🔴 2, 17 unch. · avg -0.9% · Bytecode: 🟢 1, 18 unch. · avg +1.5%
temporal.js — Interp: 6 unch. · avg -1.9% · Bytecode: 🔴 1, 5 unch. · avg -0.4%
tsv.js — Interp: 9 unch. · avg +2.4% · Bytecode: 🔴 1, 8 unch. · avg -1.4%
typed-arrays.js — Interp: 🟢 6, 🔴 2, 14 unch. · avg +13.8% · Bytecode: 🟢 2, 🔴 8, 12 unch. · avg -1.5%
uint8array-encoding.js — Interp: 🔴 5, 13 unch. · avg -11.2% · Bytecode: 🟢 6, 🔴 3, 9 unch. · avg +18.4%
weak-collections.js — Interp: 🟢 1, 🔴 7, 7 unch. · avg -14.6% · Bytecode: 🟢 2, 🔴 4, 9 unch. · avg -6.6%
Deterministic profile diffDeterministic profile diff: no significant changes. Measured on ubuntu-latest x64. Each PR run also builds the |
Contributor
test262 Conformance
Areas closest to 100%
Per-test deltas (+1 / -0)Newly passing (1):
Steady-state failures are non-blocking; regressions vs the cached main baseline (lower total pass count, or any PASS → non-PASS transition) fail the conformance gate. Measured on ubuntu-latest x64, bytecode mode. Areas grouped by the first two test262 path components; minimum 25 attempted tests, areas already at 100% excluded. Δ vs main compares against the most recent cached |
The recursive serializer built a result string per object/array level (SB.ToString) that the parent re-appended into its own buffer, re-copying the deepest content once per ancestor level — an O(depth^3) term that dominated deep stringify even after the indent fix in the previous commit. Thread a single TStringBuffer through the recursion (WriteValue/WriteObject/ WriteArray) so every value is written once; the root calls ToString once. This drops deep pretty-print from O(depth^3) to O(depth^2) (the unavoidable indentation-output size): a clean --prod A/B at depth 480 goes 583ms -> 62ms (~9.4x), per-doubling scaling falls from ~8x to ~3x. Output is byte-identical across JSON/JSON5, gap/no-gap, replacer, toJSON, RawJSON, and circular detection. Removes the now-dead StringifyValue. Add deep no-gap object/array, interleaved object/array, and JSON5 deep pretty-print regression tests, plus a depth-200 indent benchmark. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
frostney
changed the title
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
1 participant
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.
Summary
JSON.stringifypretty-printing of deeply nested values was O(depth³) from two independent terms. This PR fixes both, bringing deep pretty-print to O(depth²) — the unavoidable size of the indentation that must appear in the output.MakeIndentbuilt the indent withResult := Result + FGapin a loop (O(level²) per call, called per node). Replaced withDupeString(FGap, ALevel)— a singleSetLength+Movepass (O(level)).SB.ToStringand the parent re-appended it, re-copying the deepest content once per ancestor level. Replaced by threading one sharedTStringBufferthrough the recursion (WriteValue/WriteObject/WriteArray) so every value is written once and the root callsToStringonce.{}/[]) with no output change.StringifyValue.LengthOfArrayLikenow runs inside thetry/finally, so a throwing.lengthcoercion no longer leaks a stack entry.toJSON, RawJSON, and circular-reference detection; the stringifier's re-entrancy (FGap/stack save-restore) is preserved.Closes #811
Testing
Details:
Full suite 10956/10956 passing in both interpreted and bytecode modes.
./format.pas --check: 370 files, all formatted correctly.Byte-identical cross-check: the new framing tests pass against both the pre-rewrite and post-rewrite code (old behavior == new behavior).
New regression tests: deep nested object/array per-level indentation, multi-character gap, depth-30 indent-width scaling, deep no-gap object/array, interleaved object/array, and a deep JSON5 pretty-print (trailing comma per level). New depth-50 object/array and depth-200 object indent benchmarks; all execute.
Performance (clean
--prodA/B, depth 120/240/480, 40 reps, min of 5):Per-doubling scaling falls from ~8× (O(depth³)) to ~3× (O(depth²)).
Code review:
/code-review(4 independent finder passes) — no defects.