Set the bounding_box within the forward_transform property

[WilliamJamieson]

Astropy modeling does not attempt to determine the bounding box for compound models. Thus when the wcs.forward_transform computes the transform model by chaining all the wcs transforms together the bounding box is lost. The bounding box for the wcs.forward_transform should always be the bounding_box for the wcs. Where ever the forward_transform is actually used the bounding box is manually attached.

This PR moves that attachment inside the forward_transform property itself.

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 87.45%. Comparing base (9cd8552) to head (81790ad).
Report is 8 commits behind head on master.

Additional details and impacted files

@@            Coverage Diff             @@
##           master     #532      +/-   ##
==========================================
+ Coverage   87.42%   87.45%   +0.02%     
==========================================
  Files          22       22              
  Lines        3874     3891      +17     
==========================================
+ Hits         3387     3403      +16     
- Misses        487      488       +1     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[nden]

Can you remind me why is it not necessary to reverse the bounding box when assigning to a transform? Is the default in modeling now order=F?

[WilliamJamieson]

gwcs/gwcs/wcs.py

Lines 1282 to 1312 in c01c6d0

	def bounding_box(self):
	"""
	Return the range of acceptable values for each input axis.
	The order of the axes is `~gwcs.coordinate_frames.CoordinateFrame.axes_order`.
	"""
	frames = self.available_frames
	transform_0 = self.get_transform(frames[0], frames[1])
	try:
	bb = transform_0.bounding_box
	except NotImplementedError:
	return None
	if (
	# Check that the bounding_box was set on the instance (not a default)
	transform_0._user_bounding_box is not None
	# Check the order of that bounding_box is C
	and bb.order == "C"
	# Check that the bounding_box is not a single value
	and (isinstance(bb, CompoundBoundingBox) or len(bb) > 1)
	):
	warnings.warn(
	"The bounding_box was set in C order on the transform prior to being used in the gwcs!\n"
	"Check that you indended that ordering for the bounding_box, and consider setting it in F order.\n"
	"The bounding_box will remain meaning the same but will be converted to F order for consistency in the GWCS.",
	GwcsBoundingBoxWarning
	)
	self.bounding_box = bb.bounding_box(order="F")
	bb = self.bounding_box
	return bb

Returns a astropy.modeling.bounding_box.ModelBoundingBox not a tuple. That full object will be processed correctly by astropy.

[nden]

Regression tests have the same failures as main.

https://github.com/spacetelescope/RegressionTests/actions/runs/12168121751

[nden]

Should we be more explicit in setting the bounding_box in assign_wcs, e.g. here?