v0.30.0 — recurring parser performance and fleet measurement
Highlights
v0.30.0 improves recurring-parser performance and makes fleet measurements
clean/error-aware, resumable, revision-pinned, and fail-closed.
- Reused parsers invalidate the 16,384-entry clean-zero front cache by epoch,
reducing recurring one-byte KDL and JSON wall time by 33.10% and 35.67%
with unchanged allocation counts. - The canonical materialized full-parse, single-byte incremental, and no-edit
benchmark trio remains neutral. - Exact-blob runtime profiles retain required D and Groovy stack ceilings and
eliminate C#'s redundant first same-stack retry without changing conservative
behavior for adapted grammars or diagnostic overrides. - Real-corpus scoreboards now distinguish clean, error-bearing, and stopped
parses outside the timed path. - Resumable one-language shards carry repository revision, source cleanliness,
corpus-lock, host, configuration, classification, and hard-gate evidence;
the reducer recomputes and authenticates the combined board without rerunning
parsers. - Corpus inventory now rejects language checkouts with no eligible files rather
than allowing empty coverage to look complete.
Install with:
go get github.com/odvcencio/gotreesitter@v0.30.0See the v0.30.0 changelog for the complete change list.