ExtSlice Support

[Stefan Doerr]

I am playing around a bit with tangent after your helpful reply in reddit ;)

So I was trying to reshape an array. First I had:

  File "<stdin>", line 2
    r2 = np.sum(wrapped ** 2, axis=1)[:, np.newaxis]
                                   ^
TangentParseError: Extended Slices are not supported

Then I tried with reshape
r2 = np.sum(wrapped ** 2, axis=1).reshape(-1, 1)
and it gave: AttributeError: 'NoneType' object has no attribute 'reshape'

Then I tried

r2 = np.sum(wrapped ** 2, axis=1)
r2 = r2.reshape(-1, 1)

which gave AttributeError: module 'builtins' has no attribute 'r2'

and there I finally got the hint and did:

r2 = np.sum(wrapped ** 2, axis=1)
r2 = np.reshape(r2, (-1, 1))

Which worked. I guess it would be nice for new users to add a line to the docs not to use the array methods and instead use directly numpy functions.

[Alex Wiltschko]

Good point. We don't currently have the machinery to detect that r2 is an object with methods. I'll add a line to the docs soon making this clear.

We should support ExtSlice, though, that should have worked off the bat.

Requires supporting it in our ANF transform, probably minimal change:

tangent/tangent/anf.py

Lines 109 to 123 in 01f861a

 if isinstance(node.slice, gast.Index): 
   node.slice.value = self.trivialize(node.slice.value) 
 elif isinstance(node.slice, gast.Slice): 
   name = self.namer.name(node.slice) 
   target = gast.Name(id=name, ctx=gast.Store(), annotation=None) 
   stmt = gast.Assign(targets=[target], value=None) 
   self.prepend(stmt) 
   stmt.value = gast.Call( 
       func=gast.Name(id='slice', ctx=gast.Load(), annotation=None), 
       args=[ 
           self.trivialize(arg) if arg else 
           gast.Name(id='None', ctx=gast.Load(), annotation=None) 
           for arg in [node.slice.lower, node.slice.upper, 
                       node.slice.step]], 
       keywords=[]) 

[Bart van Merriënboer]

Should be fixed in 0ef0225.