Test: regression for iccV5DspObsToV4Dsp rXYZ/gXYZ/bXYZ tag type (#1362)

[colourbill-ctrl]

Paired regression test for #1362 (fix in #1364). In-repo branch (adds a .github/scripts regression + CTest registration, so it needs the in-repo CI per the #1363/#1365 discussion).

What it checks

Runs iccV5DspObsToV4Dsp on the committed V5 display + observer inputs and inspects the on-disk type signature of each matrix-column tag. Fails if any of rXYZ/gXYZ/bXYZ is still s15Fixed16ArrayType ('sf32', 73663332) instead of XYZType ('XYZ ', 58595a20).

Verification

Producer	Result
master (no fix)	FAIL, exit 2 (all three 'sf32')
with #1364	PASS, exit 0 (all three 'XYZ ')

Plain functional test; skips cleanly when the tool or fixtures are absent.

Files

• .github/scripts/iccdev-issue-1362-v5dspobs-xyztype-regression.sh
• .github/ci/test-data/v5dspobs-lcddisplay.icc (V5 display input), …/v5dspobs-cat8lab.icc (V5 observer input)
• Build/Cmake/Testing/CMakeLists.txt — CTest registration (iccdev.issue-1362-v5dspobs-xyztype-regression)

⚠️ Merge order: depends on the producer fix in #1364 — expected red until #1364 lands on master. Merge #1364 first, then this.

🤖 Generated with Claude Code

[xsscx]

2026-06-15 22:04:22 UTC

[xsscx]

2026-06-15 22:11:03 UTC

Rebase in Progress

• Easy Fix: iccdev_add_script_test(...) block for issue iccV5DspObsToV4Dsp writes rXYZ/gXYZ/bXYZ matrix-column tags as s15Fixed16ArrayType ('sf32') instead of XYZType ('XYZ ') #1362 is missing its closing )

In Process

[colourbill-ctrl]

Same diagnosis here — pushed the minimal fix as 63f406d.

The "Update branch" merge of master (2a69ca8) dropped the closing ) of the iccdev.issue-1362-v5dspobs-xyztype-regression iccdev_add_script_test() call, so the #1362 block ran straight into the #1361 block master contributed. CMake aborted at the top of Testing/CMakeLists.txt with Parse error. Function missing ending ")". End of file reached., which is why every build-dependent job (JSON round-trip, Coverage, Example Project, CTest ASAN+UBSAN) failed at configure before any test ran.

63f406d just restores the ) + blank separator so #1362 and #1361 are two well-formed calls again (paren balance verified). CI is re-running. If you've got a rebase staged on your side, feel free to take whichever is cleaner — flagging so we don't double-push.

[xsscx]

Doing a Rebase and will have that back up shortly unless you push and open PR first.. found macOS bug when Auto-Merge is On.

[colourbill-ctrl]

Heads-up in case GitHub's PR-head cache is hiding it on your end: the branch already carries the fix — origin/test-1362-v5dspobs-xyztype is at 63f406d right now (git ls-remote confirms), and this PR is open on it. The PR-head field lagged at 2a69ca8 for several minutes after my push, so the UI may still show the old SHA.

That said — all yours. If your rebase is cleaner (and folds in the macOS Auto-Merge fix you found), force-push right over 63f406d; the content's identical and I won't push again. Just flagging so you don't redo work that's already on the wire. Happy to look at the macOS Auto-Merge-on bug separately if it wants its own issue.

[colourbill-ctrl]

On the macOS-with-Auto-Merge bug — happy to take a swing at it, but I want to triage portable-vs-device-only before touching anything. Could you drop:

• a link to the failing macOS run/job (or the gh run view --log-failed excerpt), and
• one line on the trigger — is it the build/test failing on the macOS runner, or GitHub's Auto-Merge flow itself misbehaving (e.g. the throwaway merge commit it builds)?

Reason I ask: most of our macOS reds turn out to be portable toolchain strictness (Apple clang/libc++, char signedness, APFS case-sensitivity on include paths) — fixable here with the macOS CI runners as the verifier, no local Mac needed. If it's genuinely device-runtime-only or an Auto-Merge process quirk, that's yours and I'll stay out of it. Just don't want to guess blind.

[xsscx]

I too am seeing a major lag / latency, nothing out of the ordinary I guess.. This is ready to go, open a pr when time permits https://github.com/InternationalColorConsortium/iccDEV/tree/refs/heads/fix-rev-1362

[colourbill-ctrl]

Opened #1368 from your fix-rev-1362 branch to restore the dropped #1362 test.

Confirmed what happened: this PR is marked merged (21da415), but its auto-merge resolved the Testing/CMakeLists.txt conflict against master and dropped all four #1362 artifacts — the script, both .icc fixtures, and the iccdev_add_script_test() registration. Master right now has the #1364 producer fix (CIccTagXYZ) and the #1356/#1361 tests, but no #1362 guard. fix-rev-1362 restores exactly those four pieces, and its #1362 block carries the closing ) (paren balance verified). #1368 has the details.

[colourbill-ctrl]

Postmortem — this PR merged but its test did not land

Recording what happened for the archive, because the GitHub UI is misleading here.

Timeline (all 2026-06-15 UTC):

• 22:07 — GitHub's Update branch button merged master into this branch (merge commit 2a69ca8). It mis-resolved the Build/Cmake/Testing/CMakeLists.txt conflict and dropped the closing ) of the iccdev.issue-1362-v5dspobs-xyztype-regression iccdev_add_script_test() call, so the #1362 block ran straight into the #1361 block master had contributed. Every build-dependent job (JSON round-trip, Coverage, Example Project, CTest ASAN+UBSAN, all macOS) then failed at CMake configure with Parse error. Function missing ending ")". End of file reached. — before any test ran. The macOS reds were this same parse error, not a Mac-specific bug.
• 22:12 — Auto-merge fired and merged this PR anyway (merge commit 21da415). Resolving the same CMakeLists.txt conflict against master, it dropped all four iccV5DspObsToV4Dsp writes rXYZ/gXYZ/bXYZ matrix-column tags as s15Fixed16ArrayType ('sf32') instead of XYZType ('XYZ ') #1362 artifacts: the regression script, both .icc fixtures, and the CTest registration. Net result: the PR shows merged, but the iccV5DspObsToV4Dsp writes rXYZ/gXYZ/bXYZ matrix-column tags as s15Fixed16ArrayType ('sf32') instead of XYZType ('XYZ ') #1362 test never reached master. Master kept the Fix: iccV5DspObsToV4Dsp emits rXYZ/gXYZ/bXYZ as XYZType, not s15Fixed16ArrayType (#1362) #1364 producer fix (CIccTagXYZ) and the Fix: XML serializer corrupts zero header colour-space signature on round-trip (#1355) #1356/Add CodeQL #1355 #1361 tests, leaving the rXYZ/gXYZ/bXYZ XYZType fix with no regression guard.

Resolution: restored via #1368 (from @xsscx's fix-rev-1362 branch) — re-adds the script + two fixtures + the CTest registration, with the #1362 block's closing ) intact (paren balance verified). #1368 merged at 22:49 (00f04e1), 30 checks green. Master now carries a single, balanced #1362 registration.

Lessons:

1. GitHub's Update branch / auto-merge conflict resolution can silently corrupt or drop content in CMakeLists.txt — a dropped ) here, dropped files at the 21da415 merge.
2. A PR can read merged while its auto-merge has silently dropped newly-added files. After any conflicted auto-merge, diff the merge result against the PR head to confirm the intended files actually landed.