[N-Dekker]

Reordered the creation of itkCType objects in initialize_c_types_once, in order to fix the following two issues:

• itkCType.GetCTypeForDType(np.uint32) should return the CType for unsigned int #6782
• On Linux, GetCTypeForDType(np.uint64) should return UL, and GetCTypeForDType(np.int64) should return SL #6785

Documented that the order in which these itkCType objects are created is relevant for the mapping from NumPy dtype to itkCType.

Added a Python unit test to check the round-trip itkCType ==> dtype ==> itkCType, for numeric C types with specified sizes ("fixed width types"): int8_t, uint8_t, int16_t, uint16_t, etc.

[hjmjohnson]

This looks good. Will approve once tests pass and ghostflow is message is fixed. Curious why this is a WIP?

[dzenanz]

WIP probably because it just adds a test, not a fix.

[greptile-apps]

Greptile Summary

Summary

This change adds a round-trip test for the fixed-width ITK C type aliases exposed to Python.

The executed check loaded the dtype mapping and verified that all ten covered aliases resolve back to the identical itkCType object on both the parent revision and this change. No defects were found.

T-Rex validation blocked

The full wrapped-package test could not run because the required pixi package manager tool is not installed. A prebuilt ITK Python package and build directory were also unavailable.

Confidence Score: 5/5

No functional or security defect was found in the added fixed-width dtype round-trip coverage.

The executed source-level mapping check covered all ten aliases added by the test and observed identical round-trip behavior on both the parent revision and this change. There are no final review findings.

Files Needing Attention: No files require changes. The intended wrapped-package test remains useful to run in an environment with Pixi and a configured ITK Python build.

T-Rex Logs

What T-Rex did

• Ran a Python check that loads Wrapping/Generators/Python/itk/support/types.py and asserts object-identity dtype round trips for ten fixed-width aliases, and the check passed on both the parent revision and the PR head.
• Collected a reproducible script and before/after logs showing the round-trip results for the ten aliases, and captured both the PR head and parent revision outputs for review.
• Attempted the Pixi configure/build/CTypeTest flow to validate the built package, but the process could not start because Pixi is unavailable.
• Documented the built-package validation blocker, noting that the Pixi-based validation could not proceed due to Pixi unavailability.

_{Ran code and verified through T-Rex}

_{Reviews (1): Last reviewed commit: "ENH: Test `itkCType.GetCTypeForDType(itk..." | Re-trigger Greptile}

[N-Dekker]

This looks good. Will approve once tests pass and ghostflow is message is fixed. Curious why this is a WIP?

Thanks @hjmjohnson! Two days ago, this was still WIP, because I expected the added test to fail. And indeed, it said on Linux-Build16899-PR6783-GetCTypeForDType-round-trip-Python test CTypeTest:

FAIL: test_itk_ctype_numpy_dtype_roundtrip (__main__.CTypeTestCase.test_itk_ctype_numpy_dtype_roundtrip)
Tests round-trip from itkCType to NumPy dtype and back, for the itkCType aliases with a specified size
----------------------------------------------------------------------
Traceback (most recent call last):
  File "/home/vsts/work/1/s/Wrapping/Generators/Python/Tests/test_ctype.py", line 88, in test_itk_ctype_numpy_dtype_roundtrip
    self.assertEqual(itk.itkCType.GetCTypeForDType(itk_ctype.dtype), itk_ctype)
AssertionError: <itkCType unsigned long long> != <itkCType unsigned long>

This was fixed by the commit BUG: Let initialize_c_types_once create ULL and SLL before UL and SL, that I added to this PR.

So it was originally WIP, because I was doing Test Driven Development (TDD) 😇