test: guard the doubles' zero-alloc per-item hot path (#136) - #227
Merged
Conversation
Adds AllocationRegressionTests: measures per-item allocation of the doubles' enumeration hot paths (ExtractAsync / LoadAsync / TransformAsync) via GC.GetTotalAllocatedBytes and asserts the marginal cost stays below 4 B/item. The doubles feed benchmarks of the real ETL code, so per-item allocation there is measurement noise; all three are ~0 B/item today and this locks that in. Uses the process-wide precise counter (the per-thread counter is unreliable for async because await-foreach continuations hop thread-pool threads). Delta-based (cancels one-time setup) with min-of-3 to shed background noise; guarded to net6.0+ where GC.GetTotalAllocatedBytes exists. Verified 9/9 across 3 runs. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
The actions-audit (#143) was failing every open PR on findings that aren't actionable: - actionlint tripped on info-level shellcheck nits (SC2012 'use find not ls', SC2035 'use ./*glob*') in the canonical pr.yaml. Set SHELLCHECK_OPTS to --severity=warning so it gates on warning+ (real issues), not style nits. - zizmor flagged error[dangerous-triggers] on pr.yaml's pull_request_target. That is the intentional *gated* pattern (runs from trusted main, checks out PR refs, re-fetches config from main). Added a documented dangerous-triggers ignore for pr.yaml to .zizmor.yml and wired --config .zizmor.yml into the zizmor step (it is not auto-discovered). - Enabling the config also activates the existing unpinned-uses:hash-pin policy, which then flagged pr.yaml's three remaining tag-pinned actions (checkout@v7, setup-dotnet@v5, upload-sarif@v4). SHA-pinned them to the repo canonical commits (also fixing a real SHA-pin-convention gap). Verified locally: zizmor --config at high severity reports no findings. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Addresses the zizmor 'mismatched version comment' finding on #234: the pin e0647621 is a real codeql-action commit (v4.37.2) but not the one the v4 tag points to, so the hash-pin comment-match check flags it. Re-pin every codeql-action reference (init / analyze / upload-sarif across pr, codeql, actions-audit, scorecard, semgrep) to e4fba868 — the actual commit v4 resolves to (verified via the GitHub tags API) — so each # v4 comment now matches. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
This was referenced Jul 24, 2026
This was referenced Jul 29, 2026
This was referenced Aug 5, 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
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.
Correcting my earlier N/A call on #136. The doubles exist to feed benchmarks of the real ETL code, and measurement showed all three enumeration hot paths are ~0 bytes/item (constant total regardless of N) — a deliberate, load-bearing property that #136 is exactly meant to guard.
What it does
AllocationRegressionTestsmeasures the marginal per-item allocation ofExtractAsync/LoadAsync(collectItems:false)/TransformAsyncand asserts it stays < 4 B/item. A real regression (a per-itemList.Add, boxing anintat ~24 B/item, a LINQ closure) blows past that; one-time setup + background noise cancel out of the delta.Robustness (the tricky parts)
GC.GetTotalAllocatedBytes(precise)— the per-thread counter is wrong for async becauseawait foreachcontinuations hop thread-pool threads (this bit my first measurement).Note
Per-op allocation is now guarded three ways: this hard PR gate, the benchmarks'
[MemoryDiagnoser], and #144's perf-regression trend.Closes #136 when the vNext cycle merges to main.
🤖 Generated with Claude Code