Fix Swift recovery replay checkpoint correctness - #967
Merged
Conversation
- Replay deterministic finite automaton recovery tokens from the exact skipped-prefix offset only after span, state, and error-mode proofs succeed - Preserve complete external scanner checkpoints across failed scans and incremental fast-forward so recovery and relex use the correct scanner state - Reset recovery memo and merge scratch state at operation and pool boundaries to prevent stale state and undersized reservations from affecting later parses - Add focused parser, scanner, pooled Swift parity, and checkpoint tests; document the verified minimal witness while keeping the larger issue #576 corpus mismatches open Refs #576 Buckley-Change-Hash: sha256:b526b18c7bec50bf500aa7168627c37b3f42608df00dbe8f0b97131904899381 Buckley-Change-Stats: files=23 insertions=1184 deletions=113 binaries=0
- Restore external scanner state after failed scans by default, while supporting explicit state retention for scanners such as Swift. - Reject synthetic end-of-file relex substitutions and limit Swift operator deferral to scopes with an unmatched opening angle. - Preserve the standard recovery memo slab across temporary growth without expanding shorter retained slabs. - Add regression coverage and document the parity, correctness, and performance results for the repair. Refs #576 Buckley-Change-Hash: sha256:cd235a3257c774f7353d9f62de58b8b69dc95b19a6eb205a7175b1b8288c1db1 Buckley-Change-Stats: files=12 insertions=436 deletions=28 binaries=0
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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
Correctness
The 20-byte
let x = unsafe bar()witness now matches locked C.These large witnesses still differ from locked C:
stdlib_FloatingPointToString.swiftstdlib_CollectionAlgorithms.swiftKeep issue #576 open.
All focused Docker gates pass. They cover:
CI repair
External scanners now restore their start state after a terminal failed scan by default. Swift explicitly retains its required failed-scan state.
Generic relex rejects a synthetic end-of-file token before the source end. The Swift repair preserves a trailing
custom_operatortoken.Temporary CNode memo growth now restores an existing warm standard slab. A shorter retained slab stays short.
These exact private Docker gates pass:
TestDispatcherArmCensusA0ManifestTestSwiftOptionalGenericCloseDoesNotStarveCustomOperator/trailing_after_functionTestCNodeMemoTemporaryTierLivesForParseOperationTestMarkdownIncrementalScannerCertification/256KiB/replace/middle, with 73424/262341 bytes reusedPublic continuous integration for exact head
d5c38cd8a06be1ae2a980b0d081927bb20064bafis pending.Performance
The original 20-seed randomized benchmark geometric mean improves 3.57 percent. Bytes per operation and allocations per operation stay unchanged.
One original warmed before sample uses 594240 KiB. One original warmed after sample uses 597160 KiB. The observed increase is 0.491384 percent.
For the CI repair, the primary trio median geomean changes by +1.242879 percent across 20 randomized seeds. Each p-value exceeds 0.60. Bytes and allocations per operation stay unchanged.
The repair's warmed large-witness sample changes from 565920 KiB to 565280 KiB.