iscomplexobj(BlockArray) after #259

[tbalke]

I ran into an error when running examples/scripts/ct_astra_weighted_tv_admm.py.
It seems to be the case that in

scico/scico/functional/_norm.py

Lines 102 to 105 in 4381be5

	if snp.iscomplexobj(v):
	out = snp.exp(1j * snp.angle(v)) * tmp
	else:
	out = snp.sign(v) * tmp

the if-clause gets executed (because v is a BlockArray, see below) and the arrays get converted to complex64, causing the following output:

---------------------------------------------------------------------------
ValueError                                Traceback (most recent call last)
~/pythonModules/scico/examples/scripts/ct_astra_weighted_tv_admm.py in <module>
    123 )
    124 print(f"Solving on {device_info()}\n")
--> 125 admm_unweighted.solve()
    126 x_unweighted = postprocess(admm_unweighted.x)
    127 

~/pythonModules/scico/scico/optimize/admm.py in solve(self, callback)
    679         for self.itnum in range(self.itnum, self.itnum + self.maxiter):
    680             self.step()
--> 681             self.itstat_object.insert(self.itstat_insert_func(self))
    682             if callback:
    683                 self.timer.stop()

<string> in itstat_func(obj)

~/pythonModules/scico/scico/optimize/admm.py in norm_dual_residual(self)
    579         out = 0.0
    580         for zi, ziold, Ci in zip(self.z_list, self.z_list_old, self.C_list):
--> 581             out += norm(Ci.adj(zi - ziold)) ** 2
    582         return snp.sqrt(out)
    583 

~/pythonModules/scico/scico/_generic_operators.py in adj(self, y)
    600             return ComposedLinearOperator(self.H, y)
    601         if self.output_dtype != y.dtype:
--> 602             raise ValueError(f"dtype error: expected {self.output_dtype}, got {y.dtype}")
    603         if self.output_shape != y.shape:
    604             raise ValueError(

ValueError: dtype error: expected float32, got complex64

This is an issue that only arises after #259.

• The first difference to earlier commits is that in
  

  scico/scico/optimize/admm.py

  Line 657 in 4381be5

  zi = gi.prox(Cix + ui, 1 / rhoi, v0=zi)

  
  zi is a BlockArray, whereas in previous commits it is not. Why?
• Secondly, the handling of the snp.iscomplexobj is not the same as shown below. Is this desired?

git checkout 4ac440a (before Simplify BlockArray implementation (#259))

In [22]: a = BlockArray.array([snp.zeros(2), snp.ones(2)])

In [23]: a
Out[23]: 
scico.blockarray.BlockArray: 
DeviceArray([0., 0., 1., 1.], dtype=float32)

In [24]: snp.iscomplexobj(a)
Out[24]: False

git checkout 4381be5 (after Simplify BlockArray implementation (#259))

In [2]: a = snp.blockarray([snp.zeros(2), snp.ones(2)])

In [3]: a
Out[3]: [DeviceArray([0., 0.], dtype=float32), DeviceArray([1., 1.], dtype=float32)]

In [4]: snp.iscomplexobj(a)
Out[4]: [DeviceArray(False, dtype=bool), DeviceArray(False, dtype=bool)]

In [5]: bool(snp.iscomplexobj(a))
Out[5]: True

@bwohlberg