Back to Eliza

CTS/VTS + benchmark commit duplication audit (2026-05-18)

packages/chip/docs/notes/cts-vts-dedup-2026-05-18.md

2.0.13.1 KB
Original Source

CTS/VTS + benchmark commit duplication audit (2026-05-18)

Background

Five CTS/VTS-and-benchmark workstream commits landed on develop twice: once through ws/pd-package (where they had been committed by mistake) and again via ws/cts-vts-bench-extracted (the correct cherry-pick path). Both merges are on develop, so each subject appears twice in git log develop --oneline.

This note records the SHA pairs and confirms that the duplication is a no-op: the second copy was auto-deduplicated by git's merge resolver (identical patch, already-applied content), so no revert is needed.

SHA pairs (verified identical via git patch-id --stable)

SubjectSHA A (via ws/pd-package)SHA B (via ws/cts-vts-bench-extracted)patch-id
Add Cuttlefish riscv64 bring-up recipe1cec2293b2fceafa97e77cbd2de2b34b483dae1a321b7b1db14509
Add CTS/VTS smoke plan and fail-closed wrapper scripts9e97ec21945040f0cd50bcf51d8193b5fce267eef57ffe8a1e12b5
Add benchmark stdout parsers with unit testsb72bdbf0cce977810475cfc6bc667cbfbd9ea450304be915379044
Upgrade mobile_smoke TFLite generator to conv+relu+fc nete154598f8b573f54b87b61450c1c9e8dfe2f04c7db46beb66de22a
Add benchmark claim ladder L0-L4ce2cfaf65344686a32683658dcb478d07671ca99f29f24383411df

For every pair the git patch-id --stable hashes are byte-identical, meaning the two commits introduce exactly the same diff. Because the first copy was already applied to the tree by the time the second copy's merge ran, the merge resolver folded the second copy in as a zero-effect change. The working tree on develop carries each file exactly once.

Aside: third CTS/VTS commit (85de8d2)

git log also shows a third commit with the subject "Add CTS/VTS smoke plan and fail-closed wrapper scripts" at SHA 85de8d2, which arrived on develop via the earlier ws/security-boot-work merge. This is not a duplicate of the pair above — it is a larger, earlier draft (191/88/76 lines vs. 149/64/56 lines) that was superseded when the pair's smaller, canonical version was merged in later. The surviving file content on develop matches the 149-line variant from 9e97ec2/1945040, so 85de8d2's extra lines are not present in the tree. No action is needed; it is recorded here only so future archeology does not mistake it for a third copy of the same change.

Outcome

No revert was performed on ws/dedupe-cts-vts. The duplication is structural (two merge paths) but content-neutral (identical patches, auto-deduplicated by the merge resolver). This note is the only artifact of the dedup branch.