perf(json): scan whitespace directly to avoid per-call StringView alloc#3632
Open
mizchi wants to merge 1 commit into
Open
perf(json): scan whitespace directly to avoid per-call StringView alloc#3632mizchi wants to merge 1 commit into
mizchi wants to merge 1 commit into
Conversation
Contributor
There was a problem hiding this comment.
Pull request overview
Note
Copilot was unable to run its full agentic suite in this review.
Optimizes JSON whitespace skipping to reduce allocations in the lexer hot path by replacing regex-based matching with a direct scan over the input.
Changes:
- Replaced
lexmatchon a freshly createdStringViewwith a loop that scans ASCII whitespace directly. - Added rationale comments describing allocation reductions and UTF-16 code unit scanning semantics.
💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.
Comment on lines
124
to
131
| while ctx.offset < ctx.end_offset { | ||
| let c = ctx.input.unsafe_get(ctx.offset) | ||
| if c == ' ' || c == '\t' || c == '\r' || c == '\n' { | ||
| ctx.offset = ctx.offset + 1 | ||
| } else { | ||
| break | ||
| } | ||
| } |
Contributor
Author
There was a problem hiding this comment.
Done — offset and end are now hoisted into locals at the top of the function and ctx.offset is written once after the loop. New revision pushed.
Comment on lines
123
to
132
| fn ParseContext::lex_skip_whitespace(ctx : ParseContext) -> Unit { | ||
| let rest = ctx.input.view(start_offset=ctx.offset, end_offset=ctx.end_offset) | ||
| lexmatch rest with longest { | ||
| ("[ \t\r\n]+", next) => ctx.offset = ctx.end_offset - next.length() | ||
| _ => () | ||
| while ctx.offset < ctx.end_offset { | ||
| let c = ctx.input.unsafe_get(ctx.offset) | ||
| if c == ' ' || c == '\t' || c == '\r' || c == '\n' { | ||
| ctx.offset = ctx.offset + 1 | ||
| } else { | ||
| break | ||
| } | ||
| } | ||
| } |
Contributor
Author
There was a problem hiding this comment.
Added a focused covering empty, all-whitespace, leading-whitespace, no-whitespace, and the non-ASCII (U+00A0 NBSP must not count) cases — included in the amended commit.
mizchi
added a commit
to mizchi/pprof-mbt
that referenced
this pull request
May 26, 2026
`ParseContext::lex_skip_whitespace` constructs a fresh StringView of the remaining input on every call (to feed `lexmatch [ \t\r\n]+`), which the new memprofile flagged as the #1 alloc source in JSON parsing. Replacing it with a direct ASCII scan over the existing view drops the json_parse bench from 145.13 MB / 13.15 M allocs to 107.37 MB / 9.85 M allocs (-26.0 % / -25.1 %) at unchanged output. Lives under notes/pr-drafts/06- to follow the existing PR-draft convention. Filed upstream as moonbitlang/core#3632. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
`ParseContext::lex_skip_whitespace` is called at the start of every `lex_value` / `lex_after_*` step. The original implementation built a `StringView` of the remaining input on every call and ran `lexmatch [ \t\r\n]+` on it. That StringView wrapper showed up as the #1 allocation source in JSON parsing workloads. Since JSON whitespace is ASCII-only, we can scan UTF-16 code units directly on the existing input view with identical semantics and zero allocations. `offset` is held in a local for the duration of the scan so each consumed whitespace character is one `local.set` instead of a `struct.set` against `ctx`. Measured on a 197 KB JSON-array-of-1000-objects parsed 50 times (wasm-gc, wasmtime): Total alloc bytes : 145.13 MB -> 107.37 MB (-26.0 %) Total #allocs : 13.15 M -> 9.85 M (-25.1 %) StringView::view : 32.62 MB -> 7.44 MB (-77.2 %) String::view : 12.59 MB -> 0 (gone) Adds a focused `test "lex_skip_whitespace"` covering the empty-input, all-whitespace, leading-whitespace, no-leading-whitespace, and non-ASCII (U+00A0 NBSP must NOT count as JSON whitespace per RFC 8259) cases. Reproduction: github.com/mizchi/pprof-mbt (moon-pprof memprofile + summary --diff). Tests: moon test passes (6507 / 6507). Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
mizchi
force-pushed
the
json/lex-skip-whitespace-no-view
branch
from
e0c8af3 to
d1d9d2a
Compare
This was referenced May 26, 2026
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
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
ParseContext::lex_skip_whitespaceis called at the start of everylex_value/lex_after_*step. The original implementation built aStringViewof the remaining input on every call and ranlexmatch [ \t\r\n]+on it. ThatStringViewwrapper shows up as the #1 allocation source in JSON parsing workloads.Since JSON whitespace is ASCII-only, we can scan UTF-16 code units directly on the existing input view with identical semantics and zero allocations.
Numbers
Measured on a 197 KB JSON-array-of-1000-objects parsed 50 times under wasmtime + the wasm-gc backend, with every allocation captured via walrus instrumentation +
wasmtime::WasmBacktrace:StringView::viewString::viewThe
String::viewline disappears because the previous code path went through thelexmatchview-of-view materialization, which is no longer constructed.Reproduction tooling: mizchi/pprof-mbt (
moon-pprof memprofile+moon-pprof summary --diff).Test plan
moon test— 6506 / 6506 passmoon fmt --check— cleancmd/json_parsestill produces identical parse output (parsed=50 input_len=197001matches baseline)🤖 Generated with Claude Code