Compatibility with variable annotations (PEP 526)

[mjos]

Hi,

when trying to compile the following code:

n: int = 10

which uses PEP 526 style variable annotation (introduced in Python 3.6.0 beta 1) we get the following result from cython 0.26.1:

Error compiling Cython file:
------------------------------------------------------------
...
n: int = 10
^
------------------------------------------------------------

test.pyx:1:1: Syntax error in simple statement list

Is this a known limitation of cython that will be resolved in near future? Our code base uses a lot of variable annotations (helps documenting the code and refactoring) and this issue prevents us from using cython without code changes.

[TeamSpen210]

To allow Cython's various syntax changes, it has it's own parser which will need to be updated to handle those. Issue #1672 is also related.

[scoder]

Not currently supported, but should be trivial to implement in the parser (and ignore it there for now).

[scoder]

@mjos: could you provide a set of tests for this feature? That would simplify the implementation. Just write a .py file to put into /tests/run/ that uses different variations of variable annotations, both at the module level and inside of functions.

[mjos]

@scoder Sure, we will prepare something. Thanks.

[mjos]

@scoder I have created variable_annotations.py with a buch of examples gathered quickly from the PEP 526. Is it ok like this (see the attachment)?

variable_annotations.zip

[scoder]

Implemented in latest master. Could you please give it a try?

[mjos]

Sorry for my late reply. Works like a charm! Thanks a lot.

[lxkain]

I'm really excited about the prospect of using standard python conventions for optional static typing for cython's purposes. This is a great first step towards it! I am wondering, could we also make this work for the case of

cdef double f_c(double x):
    return x**2-x

-->

@cython.cdef
f_c(x: float) -> float:
    return x**2-x

The idea being that pure python mode completely specifies "all the things" that cython needs.

[lxkain]

Although, I am not sure how to use python's optional typing to specify the type of i in the following code snippet:

cdef int i  
for i in range(N):
  ...