Draft-quality Python 3.4 support

scipy / weave

Weave - tools for including C/C++ code within Python code.

https://pypi.org/project/scipy-weave/

Other

55 stars 27 forks source link

Draft-quality Python 3.4 support #2

Closed piannucci closed 1 year ago

piannucci commented 9 years ago

Opening a new PR extending https://github.com/scipy/weave/pull/1. Not all tests pass, but my own applications work correctly.

Open questions remain regarding how str/bytes should be handled, how file objects should be handled, and how/if the codebase can simultaneously support Python 2 and 3.

pv commented 9 years ago

FWIW, https://github.com/pv/scipy-work/commit/196829a3fbe4d41a228445331131f206fae89599, not sure if useful any more though

Some comments based on a quick browse:

Numpy takes char* == bytes, insisting on no data conversions.
Handling file objects is sort of possible to do, but it's an evil hack, and requires inserting cleanup code after the file handle is no longer needed, so it IMHO is a reasonable choice to not provide automatic conversions at all on Python 3. If needed, users can pass in the fileno() integers themselves.
Maintaining two code bases for Py2 and Py3 is usually a no-go, and usually just adding the if sys.version_info[0] >= 3 statements is enough.

rgommers commented 9 years ago

@piannucci this PR needs a rebase now. When you push an update then the TravisCI tests will run (just enabled them) in order to verify that nothing broke on Python 2.6 / 2.7.

piannucci commented 9 years ago

Okay, rebased. Also did some housekeeping. Can someone set up Travis to test Python 3.4?

pv commented 9 years ago

I think you can set up the test yourself -- edit .travis.yml and copypaste the python 2.7 section in the beginning.

piannucci commented 9 years ago

Is it a problem that through scipy.six we've added a dependency on scipy itself?

rgommers commented 9 years ago

Is it a problem that through scipy.six we've added a dependency on scipy itself?

That's the part that I just ripped out, so you put it back by accident in the rebase it looks like. The dependency isn't needed.

rgommers commented 9 years ago

The tests failed, but the log only shows the scipy build.

piannucci commented 9 years ago

My bad. Still a ton of conditional code to put in to restore Python 2 compatibility.

piannucci commented 9 years ago

Hm, that's a lot more errors than I get on my machine. GCC seems unhappy.

piannucci commented 9 years ago

Test failures on Python 3.4 down to 2 on Travis, zero on my machine.

piannucci commented 9 years ago

Though there is some spew from a blitz expression that fails to compile. Which is really two bugs: one in weave.blitz, one in the test.

piannucci commented 9 years ago

Hm, I'm seeing non-deterministic test failures.