Bulk-scan single-line string bodies#491
Open
tfoutrein wants to merge 3 commits into
Open
Conversation
`Source.__init__` built `iter([(i, TOMLChar(c)) for i, c in enumerate(self)])`, allocating one tuple and one TOMLChar per character of the whole input up front. Track an integer index into the underlying string instead: `inc()` bumps the index and reads `self[idx]`, and state save/restore snapshots the index rather than copying an iterator. Construction is O(1) and per-character work is deferred to the read. No behaviour change (full suite incl. the toml-test conformance submodule passes); ~1.07-1.14x faster parsing across document sizes.
The parser advanced one character at a time through runs of whitespace, bare-key and number characters, paying a `Source.inc()` call (attribute lookups + a `TOMLChar` build + bounds check) for every character. Add `Source.advance_while(charset)` / `advance_until(stopset)`, which scan the underlying string in a single pass and update the index and current character only once, and use them for the leading-whitespace, bare-key and number/date runs. Same value contract as the `while ... and self.inc()` loops they replace. No behaviour change (full suite incl. the toml-test conformance submodule passes; round-trip output byte-identical on a varied corpus). ~1.05-1.32x faster parsing depending on shape (e.g. ~1.26x on a poetry.lock-like file).
tfoutrein
force-pushed
the
perf/string-scan
branch
from
35d3514 to
421172e
Compare
Parsing a single-line string appended its body one character at a time (`value += current; inc()`). For long string values this dominates. Scan the run of ordinary characters up to the next delimiter, backslash or control character in a single pass (`Source.advance_until`) and append the whole slice at once; the stop character is then handled by the existing branch on the next iteration. Multiline strings keep the per-character loop (CRLF handling). The stop-set is exactly the control characters the per-character loop rejects, so InvalidControlChar / escape / delimiter handling is unchanged. No behaviour change (972 tests incl. the toml-test conformance submodule; plus a 4135-input adversarial differential — output and error-type byte-identical to the per-char loop). Up to ~5x faster parsing on string-heavy single-line documents.
tfoutrein
force-pushed
the
perf/string-scan
branch
from
421172e to
e08a656
Compare
This was referenced Jun 6, 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
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.
What
Parsing a single-line string appended its body one character at a time (
value += current; inc()). For long string values (the bulk of most config / lockfile content) this dominates.This scans the run of ordinary characters up to the next delimiter, backslash or control character in a single pass (
Source.advance_until) and appends the whole slice at once; the stop character is then handled by the existing branch on the next iteration. Multiline strings keep the per-character loop (CRLF handling).The stop-set is exactly the control characters the per-character loop rejects, so
InvalidControlChar/ escape / delimiter handling is unchanged, and a mid-string EOF raisesUnexpectedEofErrorjust as the per-charinc(exception=...)did.Benchmarks
Parsing speedup across document shapes (median, interleaved A/B vs
master, includes #489+#490):No regression on any shape (multiline-heavy and nested docs are unchanged).
Tests
Full suite passes (972 tests, incl. the toml-test conformance submodule). On top of that, a 4135-input adversarial differential (random escapes valid+invalid, every control byte 0x00–0x1F+DEL in basic & literal, unicode/astral/combining, the other-quote char inside strings, truncated/malformed inputs for error parity) is byte-identical in output and exception type to the per-character loop. No public API or behaviour change.