Commit Graph

2 Commits

Author SHA1 Message Date
7add6f0087 perf(parsing): measure scanStep() overhead, fix a real but partial cause
Add StreamingScannerPerformanceTests (issue #4), gated behind
PUREYAML_STREAMING_PERF_MEASURE=1, measuring scanStep() against scanAll() on
the largest real-world fixture (9.5MB) across a spread of chunk sizes.

Small chunk sizes (16-256 scalars) show a real, chunk-size-dependent rescan
cost, up to 6x at 16, consistent with the checkpoint-and-retry design
rescanning a fetch attempt once per chunk it spans. A control run with
isFinal: true from construction, where ranDry can never be set and no
rollback ever happens, still cost ~1.5x over scanAll(), proving a separate,
chunk-size-independent cost dominates at any realistic chunk size (4KB+): the
checkpoint was being copied on every fetch even when it could never be used.

Fixed by skipping the checkpoint entirely when reader.isFinal is true.
Re-measured rather than assumed fixed: the isolated control improved from
1.49x to 1.38x, a real but modest gain, while the large-chunk-size ratios
were unchanged within measurement noise. The remaining floor's cause is not
identified and is left open on issue #4 rather than patched further on
speculation; closing it needs a profiler, not another guess.

Whole-suite conformance (402/402) and the chunk-size differential proof both
still pass unchanged.
2026-07-10 00:14:19 +02:00
18fdf9b69a feat(parsing): add a resumable scanner increment and broaden the mutation suite
Add Scanner.scanStep(), the first increment of true chunked input streaming
(#4): Reader gains isFinal/append(_:)/markFinal(), and peek(_:) flags ranDry
instead of asserting end of input when the buffer runs out mid-fetch and more
chunks may still arrive. scanStep() snapshots the scanner before each
fetchNextToken() and restores it whenever a fetch runs dry, so a construct
spanning a chunk boundary is retried after more input lands rather than
misread as an early end of input. Every existing call site keeps the default
isFinal: true, so scanAll() and batch parsing are unchanged; ranDry can never
be set when isFinal is true, by construction. A token already queued is not
always safe to release, since libyaml's simple-key mechanism can still splice
a KEY token in front of it, so scanStep() ports libyaml's
yaml_parser_fetch_more_tokens release condition.

Proof: StreamingScannerDifferentialTests.swift drives scanStep() at several
chunk sizes over representative cases always, and over the entire
yaml-test-suite manifest (402 cases) behind PUREYAML_RUN_YAML_TEST_SUITE=1,
asserting the token sequence matches scanAll() exactly. That proof caught a
real bug: lastTokenForbidsInlineSequence() read indentEngine.tokens.last,
which goes empty once tokens are released incrementally, silently breaking
the key: - a rejection. Fixed by tracking IndentEngine.lastToken explicitly,
kept in sync with tokens.last so batch behavior (which never dequeues) is
unchanged.

Not yet wired into PureYAMLStreaming; scanStep() and its supporting types stay
internal until that integration is designed.

Also broaden the real-seed mutation property suite (#6) with the three
mutation classes it named but didn't yet cover: invalidControlCharacter,
mixedTabSpaceIndentation, and truncatedFlowCollection. The test now asserts
diagnostic quality, not just pass/fail: line must fall within the mutated
document, and column, when present, must be 1-based (some ParseError cases,
like tabIndentation, are line-only by design).
2026-07-09 23:38:39 +02:00