ENH: Add dtype argument to Cifti2Image

[effigies]

To bring into line with other Analyze-like formats.

Given that the subject that brought us here was inappropriate use of image dtypes, I am also constraining the dtypes to those listed in the standard:

The datatype field in the NIfTI header can be any of float, double, and signed or unsigned int8,
int16, int32, or int64, with bitpix set correspondingly, per the NIfTI standard. Composite
datatypes, such as complex or RGB, and bitwise storage are not supported in CIFTI. Multiple
scalar maps, or separate files, can be used instead of composite datatypes.

I also dropped float128, which is not specifically restricted, but nor is it specifically permitted. Seems unlikely to be a good idea.

[effigies]

Tests forthcoming.

[codecov]

Codecov Report

Merging #1111 (7933fae) into master (4bd0027) will increase coverage by 0.01%.
The diff coverage is 100.00%.

@@            Coverage Diff             @@
##           master    #1111      +/-   ##
==========================================
+ Coverage   92.26%   92.28%   +0.01%     
==========================================
  Files         100      100              
  Lines       12269    12285      +16     
  Branches     2399     2402       +3     
==========================================
+ Hits        11320    11337      +17     
  Misses        624      624              
+ Partials      325      324       -1     

Impacted Files	Coverage Δ
nibabel/analyze.py	98.58% <100.00%> (+0.03%)	⬆️
nibabel/cifti2/cifti2.py	96.85% <100.00%> (+0.20%)	⬆️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 4bd0027...7933fae. Read the comment docs.

[effigies]

Couple notes for a reviewer.

[effigies]

Possibly there's a better name. _Cifti2AsNiftiHeader exists to do some validation during parsing, but isn't the type of nifti_header.

nibabel/nibabel/cifti2/parse_cifti2.py

Lines 78 to 119 in 7933fae

	class _Cifti2AsNiftiHeader(Nifti2Header):
	""" Class for Cifti2 header extension """
	@classmethod
	def _valid_intent_code(klass, intent_code):
	""" Return True if `intent_code` matches our class `klass`
	"""
	return intent_code >= 3000 and intent_code < 3100
	@classmethod
	def may_contain_header(klass, binaryblock):
	if not super(_Cifti2AsNiftiHeader, klass).may_contain_header(binaryblock):
	return False
	hdr = klass(binaryblock=binaryblock[:klass.sizeof_hdr])
	return klass._valid_intent_code(hdr.get_intent('code')[0])
	@staticmethod
	def _chk_qfac(hdr, fix=False):
	# Allow qfac of 0 without complaint for CIFTI-2
	rep = Report(HeaderDataError)
	if hdr['pixdim'][0] in (-1, 0, 1):
	return hdr, rep
	rep.problem_level = 20
	rep.problem_msg = 'pixdim[0] (qfac) should be 1 (default) or 0 or -1'
	if fix:
	hdr['pixdim'][0] = 1
	rep.fix_msg = 'setting qfac to 1'
	return hdr, rep
	@staticmethod
	def _chk_pixdims(hdr, fix=False):
	rep = Report(HeaderDataError)
	pixdims = hdr['pixdim']
	spat_dims = pixdims[1:4]
	if not np.any(spat_dims < 0):
	return hdr, rep
	rep.problem_level = 35
	rep.problem_msg = 'pixdim[1,2,3] should be zero or positive'
	if fix:
	hdr['pixdim'][1:4] = np.abs(spat_dims)
	rep.fix_msg = 'setting to abs of pixdim values'
	return hdr, rep

[effigies]

This subclass/superclass swap occurs because not all spatial images have dtype overriding.

[effigies]

If anybody wants to review, please let me know by end of day (5pm UTC-4). Otherwise I'll merge and aim to release 4.0.0rc0 on Monday.