[ajitpratap0]

Summary

• TestFormatASTExpressionLinearGrowth (added in fix(formatter,transform): complete lower-case output, keep comments by GO, handle TOP/FETCH #870) flakes in the Race Detector job: left_deep_binary/lower measured 65.24x for 16x input against a 64x bound.
• Without race instrumentation the lower-case path grows linearly (n=8192→131072: 19ms→218ms, ~11x for 16x input), so this is instrumentation skew, not an O(n²) regression.
• Skip the ratio check when built with -race, following the parser's existing raceEnabled build-tag pattern. The non-race test matrix (Linux/macOS/Windows) still runs it.

Test Plan

• go test -race -run TestFormatASTExpressionLinearGrowth ./pkg/formatter/ → SKIP
• go test -count=3 -run TestFormatASTExpressionLinearGrowth ./pkg/formatter/ → PASS
• go vet ./pkg/formatter/

Breaking Changes

None (test-only).

🤖 Generated with Claude Code

https://claude.ai/code/session_01DnbRnTFErhoMJdAjVEeCg2

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

1 Skipped Deployment

Project	Deployment	Actions	Updated
gosqlx-website	Ignored	Preview	Oct 28, 2026 4:24am UTC

[github-actions]

Claude finished @ajitpratap0's task in 13s —— View job

---

Review of PR #875

Verdict: LGTM. The change is small, correct and low-risk. I read the diff and did not run the tests myself.

What I checked

• The race / !race build-tag pair is mutually exclusive, so raceEnabled is always defined exactly once.
• Both files are _test.go and use package formatter_test, which matches the package of expression_scaling_test.go. Nothing ships in the production build.
• The skip sits at the top of TestFormatASTExpressionLinearGrowth, before any work is done, so the race job wastes no time on it.
• It follows the existing parser pattern (performance_regression_race.go and performance_regression_norace.go). One small difference: the parser files are not _test.go files, while these are.

Minor, non-blocking notes

1. Coverage under race: the race job no longer checks this test at all. That is a reasonable trade, since the Linux, macOS and Windows non-race jobs still run it. A real O(n²) regression would still be caught there.
2. Flake risk elsewhere: the 64x bound is still tight for noisy runners. Your measurement of about 11x for 16x input (65.24x for 16x under race) shows plenty of headroom without race. If it ever flakes without race, consider taking the best of a few runs.
3. Naming: race_enabled_*_test.go is clear. Other packages may later want the same helper. If a third package needs it, consider moving it to a shared internal test helper. That isn't needed now.

No bugs, security concerns or performance concerns. Safe to merge once CI is green.