[hjmjohnson]

Fixes #7025: itkIndent.h fails to compile when it is included before itkMacro.h. This restores @blowekamp's May fix (commits 79f01ca, 4f524e8: itkIndent.h includes no itkMacro.h, so itkMacro.h → itkPrintHelper.h → itkIndent.h cannot loop) and adds a loud error so the cycle cannot come back silently. Supersedes #7026 by @mymonin, whose issue, diagnosis, and include-first regression check this keeps.

• itkIndent.h includes only the leaf headers ITKCommonExport.h and itkConfigure.h, and marks the legacy New()/Delete() with [[deprecated]] instead of itkLegacyMacro (which lives in itkMacro.h), with a comment saying why.
• itkPrintHelper.h stops with a message naming the fix if itk::Indent is incomplete when it is reached, which is what including itkMacro.h from itkIndent.h again would cause.
• itkPrintHelperGTest.cxx includes itkIndent.h first so every build exercises the order from COMP: itkMacro.h must be included before itkIndent.h #7025.

One commit, three files, +15/−2. No forward declarations are needed, so nothing duplicates Indent's operator<< signature.

How the cycle came back

• 79f01ca/4f524e8 (May 2026) removed #include "itkMacro.h" from itkIndent.h; an #ifndef itkMacro_h #error was tried first and dropped because it broke standalone inclusion in ITKCommonHeaderTest1.
• 70ddeb8 (Sep 2026) wrapped Indent::New()/Delete() in itkLegacyMacro under #ifndef ITK_LEGACY_REMOVE. Both come from itkMacro.h/itkConfigure.h, so itkIndent.h depended on itkMacro.h again without including it. Most files include itkMacro.h first, so only the ARMBUILD header-test chunk that happened to list itkIndent.h first failed (ENH: Eliminate const_cast from itk::ImageIterator #7021, ENH: Name the series in the non-uniform sampling warning of ImageSeriesReader #7024).
• [[deprecated]] differs from itkLegacyMacro only in that ITK_LEGACY_SILENT/ITK_LEGACY_TEST no longer silence it for these two members; nothing in ITK calls them, and both are removed under ITK_LEGACY_REMOVE.
• Including itkMacro.h from itkIndent.h (COMP: Fix missing direct include for macro definition #7026) re-creates the cycle and needs forward declarations of Indent and its operator<< in itkPrintHelper.h. That copy omits ITKCommon_EXPORT, and keeping two declarations in sync is the maintenance cost this PR avoids.

Testing

macOS arm64, Apple clang, Debug, ITKCommon + ITKImageIntensity with MAXIMUM_NUMBER_OF_HEADERS=1 (every header compiled on its own, including itkIndent.h alone):

• Build: 0 failures, 0 compiler warnings. ctest: 1013/1013 passed.
• ITK_LEGACY_REMOVE defined: itkIndent.h alone and itkIndent.cxx compile.
• Include graph (every quoted #include in ITKCommon, include guards ignored): itkIndent.h includes no other ITKCommon header (only the generated, include-free ITKCommonExport.h and itkConfigure.h), so no cycle passes through it. The same check on COMP: Fix missing direct include for macro definition #7026 finds itkIndent.h → itkMacro.h → itkPrintHelper.h → itkIndent.h.
• Regression check: adding #include "itkMacro.h" back to itkIndent.h makes itkPrintHelperGTest.cxx stop with itkPrintHelper.h:25: error: "itk::Indent incomplete: itkIndent.h must not include itkMacro.h (keep ITKCommonExport.h, itkConfigure.h)".
• pre-commit run --all-files passes.

Not tested locally: Windows, Linux, Python wrapping.

AI assistance

Claude Code traced the history, wrote the change and the compile checks, and drafted this text at @hjmjohnson's direction; the build and tests above were run locally.

[greptile-apps]

Confidence Score: 5/5

[Medium impact] This PR appears safe to merge.

Summary

This PR makes itkIndent.h work without first including itkMacro.h.

• itkIndent.h can be included before itkMacro.h without restarting the include cycle.

Diagram

%%{init: {'theme': 'neutral'}}%%
flowchart LR
  Macro["itkMacro.h"] --> Print["itkPrintHelper.h"]
  Print --> Indent["itkIndent.h"]
  Indent --> Export["ITKCommonExport.h"]
  Indent --> Configure["itkConfigure.h"]

Loading

_{Reviews (2) · Last reviewed commit: "COMP: Keep itkIndent.h free of itkMacro...." · Reviewed by Greptile}

[hjmjohnson]

Sorry for the noise here. 3 failed attempts accidently go co-mingled and pushed and did not represent what I have on my local build. I assumed my local build was the same as the PR incorrectly.

[blowekamp]

I don't think adding another header into this complicate dependency order is the correct solution. The additional header will add more complication when similar issues are encountered.

My recommendation is to keep the legacy macro where it is, but change itkIndent to directly use [deprecated note] with a comment why the macro is not used.

... this may have just happened with the latest push?

[hjmjohnson]

@blowekamp I had already done that. That was the mistake I made in commingling solutions. You must have reviewed just before the fix. Sorry.

[mymonin]

Calling [[deprecated]] directly instead of the macro works as well. Looks good to me.

[hjmjohnson]

ITK.Linux.Python on 3b548bd was canceled by Azure at 1h51m while still compiling (no errors in the log; successful runs of this pipeline take 3–4 h), so it is not a result for this change. Re-running it.

[hjmjohnson]

/azp run ITK.Linux.Python

[dzenanz]

Looks good on a glance.

[dzenanz]

Let's merge either this or #7026 today.

I still don't know why the problem started appearing yesterday. #7024 is a minimal change, which does not touch the failing test.

[hjmjohnson]

Downstream regression check: no regressions attributable to this PR in the ITK forest testbed (upstream main 3f19582 + this PR's head 3b548bd, macOS arm64, Clang). Every consumer that built before still builds, and no build or test log contains the new #error on the include cycle or any itkLegacyMacro / Indent failure.

Consumer matrix

Target	Build	Tests
ITK	PASS	17/17 ITKCommon (PrintHelper gtests incl. the new include-order test, header test)
elastix	PASS	159/164
SimpleITK	PASS	no tests registered
SphinxExamples	PASS	469/565
ANTs	PASS	179/179
BRAINSTools	PASS	295/416
OpenIGTLink	PASS	63/63
Slicer	PASS	not reported by harness
OpenIGTLinkIO, vtkAddon, IGSIO, PlusLib	PASS	same shortfalls as prior sweeps
RTK, Cleaver, PerformanceBenchmarking, SimpleITKFilters, TractographyTRX, VkFFTBackend, Impact	built	ctest skipped, see below

The sub-100% pass counts match earlier sweeps against plain main; none of their failing test logs mention Indent or itkMacro.

Harness caveats (not caused by this PR)

• The seven remote modules above built cleanly but the harness skipped their ctest because of stale IsotropicWavelets/TubeTK wrapping leftovers in the ITK build tree from a previous session. import itk from that build works (6.0.0). This is a testbed artifact, not a PR effect.
• SlicerExtensions was interrupted by hand before completion and is untested here.