fix(#1982): honour the D2B/B2D opt-out for dual-PCS profiles

[colourbill-ctrl]

Summary

Harvests the CIccXform::Create half of ci-qa-issue-1982 (ea41af37, @xsscx) and adds
a two-direction regression CTest. Part of #1982 — items 1 and 2; item 3 is already
implemented, and one behaviour question is left open for a maintainer (both below).
Deliberately not a closing keyword.

CIccXform::Create selected a spectral DToBx/BToDx transform whenever the profile
carried a spectralPCS, ignoring the caller's bUseD2BTags opt-out. Every other
space-selection site already tests !pcs as well — CIccCmm::AddXform at
IccCmm.cpp:9093/:9112 and CIccNamedColorCmm::AddXform at :11718/:11796/:11813
all read (bUseD2BxB2DxTags || !m_Header.pcs). For a profile carrying both a colorimetric
and a spectral PCS the sites disagreed: AddXform recorded the 3-sample XYZ connection
while Create returned a transform emitting 36 spectral samples, and CIccCmm::Begin()
rejected the chain.

Both Create lines were introduced together in 92ffea98 (#1018, hybrid spectral
profiles). The AddXform half already carried the !pcs term; the two Create sites did
not. With this change all seven sites agree.

Measured

Clang 21.1.3 Release, master f78e915a, Testing/SpecRef/SixChanInputRef.xml
(AToB3 6→3, DToB3 6→36, PCS=XYZ, SpectralPCS=rs0024):

iccApplyToLink link.icc 0 2 1 T 0 1 1 0 SixChanInputRef.icc 11
  before: exit 255, "status 2: Invalid space link"
  after:  exit 0,   PCS XYZData, "MPE Element Chain: 1 elements, 6->3 channels"

Same for intent 13. Both decode to icXformLutColor with bUseD2BxB2DxTags cleared.

Instrumented master, whole 222-test suite, one probe per decision site:

IN_FLIP tagIntent=1 lut=0            <- SixChanInputRef @ 11
ADDXFORM_MISMATCH src=6/6 dst=36/3   <- xform emits 36, CMM recorded 3

Why both sites in one commit

The two mis-selections cancel out. A chain whose source and destination are dual-PCS
resolves today because both ends silently take the spectral route. With only the input
site corrected:

SixChanInputRef 11 -> SixChanCameraRef 11
  master:          exit 0
  input site only: exit 255, "status 13: Unsupported PCS Link used"

#1018 is not regressed

CMYK_Hybrid_Profile's spectral sub-profile has an empty <PCS>, so !pcs holds and its
DToBx path stays open. iccdev.hybrid-pipeline passes: 43s Release, 494s ASan+UBSan.

The test

iccdev.issue-1982-dual-pcs-selection-regression, both directions.

Part 1 (source) asserts exit status and link shape. Part 2 (destination) cannot: the
chain builds a link either way, just the wrong one, so an exit-status check would be
vacuous. It builds the same chain twice — both ends opting out (11,11) and both opting
in (1,1) — and requires the two links to differ. On master they are byte-identical once
Creation Date / Profile ID / size / banner are stripped. Both arms succeed in both
builds, so this asserts the route and not an error.

Red/green:

tree	result
master f78e915a	FAIL (part 1)
input site only	FAIL (part 2)
both sites	PASS

No pre-existing test exercised the opt-out path at all. The only other script using a
tens-digit-1 intent code is iccdev-applytolink-nan-sixchan-debug.sh, and it feeds nan
as the range so it fails before AddXform. A green suite is therefore not the evidence
here — the three-way red/green above is. The test exits 77 with SKIP_RETURN_CODE when
the tools are absent, so a build that cannot run it reports Skipped rather than a green
pass.

Pre-flight

Base d1a7d42a. Dispatched ci-pr-action with ci_scope=source before opening this PR
per the suggested workflow: run
32743042807,
success, 12 jobs green / 3 skipped by scope — GCC 15.2 Strict Release LTO, Windows
MSVC + ClangCL, MinGW UCRT64, macOS Release + Debug, Tool Smoke Tests ASAN+UBSAN. On the
ASAN leg the new test ran as 149/220 Test #150 ... Passed 0.54 sec (100% tests passed, 0 failed out of 220); script CTests are not registered on the Windows leg, which
nonetheless configured the new registration under MSVC (129 tests, 100%).

Local lanes:

lane	result
Clang 21.1.3 Release, -Wall -Wextra -Wpedantic -Werror, strict warnings ENABLED	0 warnings, ctest 223/224
GCC 15.2.0 in ghcr.io/internationalcolorconsortium/iccdev-ci-regression:latest, same flags + LTO, strict warnings ENABLED	0 warnings, build and build-test-binaries both rc=0
Clang ASan+UBSan Debug	ctest 222/223, new test passes in 0.34s
detect_leaks=1 over the six affected iccApplyToLink paths, including the newly reachable ProfileMissingTag refusal	no leaks, no findings (LeakSanitizer confirmed live against a deliberately leaky control)
shellcheck 0.9.0	rc=0

The one ctest failure in each run is iccdev.spectral-tiff-preview, which needs the
imagecodecs Python module and fails identically on pristine master.

One behaviour change worth a maintainer's eye

Testing/Calc/CameraModel.xml is dual-PCS with A2B3 + B2D3 and no B2A3. As a
destination under the opt-out it now returns icCmmStatProfileMissingTag where master
silently used the spectral B2D3. A sweep of all 221 tracked XML found only three
dual-PCS profiles carrying B2D tags, and CameraModel is used only for XML→ICC, dump and
writer-failure tests — never as a CMM destination — so nothing in the suite covers it
either way. Whether an opt-out should fall back to the spectral tag when no colorimetric
tag exists in that direction is a spec call, raised on #1982 rather than decided here.

Not harvested: the branch's third change

ci-qa-issue-1982 also adds a sample-count guard in CIccCmm::AddXform (issue item 3).
Left out, measured against the branch applied verbatim:

1. CIccCmm::Begin() already does it. Its two GetNumSrcSamples()/GetNumDstSamples()
   comparisons against the CMM's counts carry the comment "Otherwise we'll have a heap
   overflow during Apply." That is exactly why the unfixed dual-PCS case fails closed with
   icCmmStatBadSpaceLink rather than overflowing.

2. It fires twice in the 222-test suite and both tests still print PASS.

   test	master	with the guard
   iccdev.issue-1336-applytolink-pcc-regression	fails on the second profile (fuzz-abst-a044ff69.icc, status 2)	fails on the first (fuzz-mntr-ef352586.icc, status 4)
   iccdev.issue-1985-encoding-lumamtx-uaf	status 2	status 4

   CI c0d0fc2: Indirect Leak in CIccProfile::ReadBasic() at IccProfile.cpp:1363 #1336 exists to prove a -PCC profile is released on the AddXform error path. Moving
   the failure to the first profile means that path is never reached, so the test goes
   vacuous while still reporting [PASS] — it only greps for a LeakSanitizer report. CIccDefaultEncProfileConverter::ConvertFromParams frees a borrowed colorEncodingParams element — heap use-after-free write, then double free (IccEncoding.cpp:259) #1985's
   own header documents the expected outcome as "the ordinary 'invalid space link'
   rejection", which the guard changes.

3. It is not a channel-count check in practice. icGetSpaceSamples() returns 0 for any
   signature it does not map, so the guard is really "reject any profile whose PCS signature
   is unrecognised or zero". Both firings were .../0 cases — nDstSpace=00000000, and
   nSrcSpace=F6BD2F3E from the fuzzed abstract profile — not genuine disagreements.

Reproduce: apply the guard and run ctest -R 'issue-1336|issue-1985'.

Checklist

• Built locally with the documented build flow in docs/build.md
• Ran relevant CTest/profile tests from docs/ctest.md
• Added or updated regression coverage for behavior changes
• Checked sanitizer coverage for memory-safety or parser changes
• Updated documentation for user-visible behavior changes
• Did not change maintainer-owned workflow, CTest, CPack, sanitizer, release, or security infrastructure unless requested by an iccDEV maintainer
• New source files include the ICC copyright and BSD 3-Clause license header
• Code style matches nearby code: 2-space indent, K&R braces, m_ members

[xsscx]

2026-08-24 15:24:40 UTC

[xsscx]

PR Note

2026-08-24 15:28:42 UTC

• Running the Copilot Cloud Review Agent to check Reporting & Coverage

[Copilot]

Pull request overview

Fixes dual-PCS transform selection to honor D2B/B2D opt-outs and adds regression coverage.

Changes:

• Updates source and destination transform selection.
• Registers the issue-1982 CTest regression.
• Adds two-direction route validation.

Reviewed changes

Copilot reviewed 3 out of 3 changed files in this pull request and generated no comments.

File	Description
IccProfLib/IccCmm.cpp	Honors D2B/B2D opt-outs for dual-PCS profiles.
Build/Cmake/Testing/CMakeLists.txt	Registers the regression test.
.github/scripts/iccdev-issue-1982-dual-pcs-selection-regression.sh	Validates colorimetric and spectral route selection.

Suppressed comments (1)

.github/scripts/iccdev-issue-1982-dual-pcs-selection-regression.sh:135

• The CTest harness sets ASAN_OPTIONS/UBSAN_OPTIONS with halt_on_error=0, so a sanitizer finding during this iccFromXml invocation can be printed while the command still returns zero. Unlike the later iccApplyToLink calls, this log is never passed to sanitizer_check (and the same gap exists for the dump logs), so the new ASAN-labeled regression can pass with a sanitizer finding in its setup/assertion tools. Check each generated-tool log before relying on its assertions.

"$from_xml" "$src_xml" "$src_icc" >"$outdir/fromxml-input.log" 2>&1 ||
  fail "iccFromXml failed on $src_xml; see $outdir/fromxml-input.log"
"$dump_profile" -v 100 "$src_icc" ALL >"$outdir/input.log" 2>&1

[xsscx]

Pre Merge Report

2026-08-24 15:40:48 UTC

TODO as Punch List following Merge

Suppressed comments (1)
.github/scripts/iccdev-issue-1982-dual-pcs-selection-regression.sh:135

The CTest harness sets ASAN_OPTIONS/UBSAN_OPTIONS with halt_on_error=0, so a sanitizer finding during this iccFromXml invocation can be printed while the command still returns zero. Unlike the later iccApplyToLink calls, this log is never passed to sanitizer_check (and the same gap exists for the dump logs), so the new ASAN-labeled regression can pass with a sanitizer finding in its setup/assertion tools. Check each generated-tool log before relying on its assertions.
"$from_xml" "$src_xml" "$src_icc" >"$outdir/fromxml-input.log" 2>&1 ||
fail "iccFromXml failed on $src_xml; see $outdir/fromxml-input.log"
"$dump_profile" -v 100 "$src_icc" ALL >"$outdir/input.log" 2>&1