BUG: SWIG typechecks must reject foreign wrapped types (fixes IsInsideBuffer, #5310)

[hjmjohnson]

Fix for Discourse #7495 and supersedes #5310. SWIG's overload dispatcher was always selecting the integer-only Index overload of ImageFunction::IsInsideBuffer because the typecheck typemaps in pyBase.i claimed viability for any object that satisfied PySequence_Check + length == dim — including wrapped ContinuousIndex / Point instances (they expose __getitem__ / __len__).

Root cause walk-through

itk::ImageFunction::IsInsideBuffer has three overloads (IndexType, ContinuousIndexType, PointType). The SWIG-generated dispatcher tries them in declaration order:

// Index typecheck (DECL_PYTHON_SEQ_TYPEMAP):
if (SWIG_ConvertPtr(arg, &ptr, SWIGTYPE_p_itkIndex3, 0) == -1
    && (!PySequence_Check(arg) || PyObject_Length(arg) != 3)
    && !PyInt_Check(arg))   _v = 0;
else                        _v = 1;

For interp.IsInsideBuffer(itk.ContinuousIndex[itk.D, 3]()):

• SWIG_ConvertPtr to itkIndex3 returns -1 (different SWIG type).
• PySequence_Check returns True (wrapped fixed-length classes expose __getitem__ / __len__).
• PyObject_Length returns 3.

⇒ _v = 1, the dispatcher commits to the Index overload, then the in typemap demands integer elements and emits

ValueError: Expecting a sequence of int (or long)

The ContinuousIndex and Point overloads were never reached. The same logic broke interp.IsInsideBuffer(itk.Point[itk.D, 3]()) and interp.IsInsideBuffer([1.5, 2.5, 3.5]).

Fix

Tighten the four affected typecheck typemaps in Wrapping/Generators/Python/PyBase/pyBase.i:

1. If SWIG_ConvertPtr succeeds for our descriptor → accept (strong type-correct match).
2. Otherwise, if SWIG_Python_GetSwigThis returns non-NULL, the input is a SWIG-wrapped instance of an unrelated type → reject, so the dispatcher keeps looking for the right overload.
3. Otherwise, fall through to the cheap scalar / sequence-length checks as before. For the integer-only SEQ typemap (Index / Size / Offset), additionally peek at element 0 and reject sequences whose first element is not a Python int — so [1.5, 2.5, 3.5] no longer claims viability for the Index overload.

This applies to:

• DECL_PYTHON_VEC_TYPEMAP — Vector, CovariantVector, Point, ContinuousIndex, FixedArray
• DECL_PYTHON_SEQ_TYPEMAP — Index, Size, Offset
• DECL_PYTHON_VARLEN_SEQ_TYPEMAP — Array, RGBPixel, RGBAPixel

Verification (hand-traced before the fix builds)

After the fix, the dispatcher resolves each test case like this:

Input	Index typecheck	ContinuousIndex typecheck	Point typecheck (fallback)	Selected overload
[1, 2, 3] (list of ints)	✅ (elem 0 is int)	—	—	Index
itk.ContinuousIndex instance	❌ (GetSwigThis non-NULL, foreign type)	✅ (ConvertPtr succeeds)	—	ContinuousIndex
itk.Point instance	❌ (foreign type)	❌ (foreign type)	✅ (final fallback)	Point
[1.5, 2.5, 3.5] (list of floats)	❌ (elem 0 is float)	✅ (sequence accepts float)	—	ContinuousIndex

Test itkLinearImageFunctionWrappingTest.py asserts each of these calls returns True for in-buffer coordinates and False for out-of-buffer coordinates.

Closes #5310.

[hjmjohnson]

Closing — content moved into #5310 (force-pushed bdbe9ae6c3 to dzenanz/isInsideBuffer). All discussion now happens on the original PR.