tamada / fauxrest

84%

main: 86%

LAST BUILD ON BRANCH fix/validate-filter-regex

branch: fix/validate-filter-regex

CHANGE BRANCH

x

Reset

Sync Branches

• fix/validate-filter-regex
• ci/force_fmt_in_CI
• docs/function-comments
• feature/overwrite-option
• feature/static-file-copy
• feature/trim-derive-type
• fix/cli-options-override
• fix/docs-deploy-workflow
• fix/publish-cargo-lock-sync
• fix/string-ordering-comparison
• main
• releases/v0.0.1
• releases/v0.0.2
• releases/v0.0.3
• releases/v0.0.4

Build # 30504752987

Build Type

push

github

Committed by tamada

Commit Message

bench: add a criterion benchmark for the $filter evaluation loop

testdata/bench measures the regex-caching change end to end through the CLI,
which is the figure a user feels but also the noisier one: it includes JSON
parsing and file writes, and it needs a generated dataset. This adds the
complementary microbenchmark over the evaluation loop, which needs no data
files and gives a repeatable regression signal.

`FilterCondition::apply` is public, so the benchmark drives the exact hot path.
Two cases run the same 10000 records through four conditions: the regex set
from testdata/bench/_config.json, and an `eq`/`exists` set as a floor for what
the loop costs with no pattern matching at all.

Verified that it actually catches the regression by running it against the
pre-fix sources: 405 ms without the cache versus 2.3 ms with it, about 175x,
while the `eq` baseline stays at 0.6-0.8 ms either way. That baseline is the
point of including it — it shows the difference belongs to the regex path and
not to the loop.

criterion goes in a new `[dev-dependencies]` section with `harness = false`,
since the built-in `#[bench]` attribute is still nightly-only.

`tempfile` stays in `[dependencies]`: `SqliteSerializer::serialize` uses
`NamedTempFile` at runtime, not just in tests.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

Coverage Stats

1244 of 1479 relevant lines covered (84.11%)

14.61 hits per line