Appveyor

[kwgoodman]

Bottleneck seems to only get tested on windows right before a release---long after a commit that breaks bottleneck on windows might have been made. See, for example, #129.

It would be nice to test bottleneck on windows with Appveyor.

Anyone up for the challenge? @Midnighter?

We could skip the flake8 and sdist tests if that is helpful (we already do those tests on Travis). Maybe windows has a 32 bit option. Travis only has 64-bit OSes.

[Midnighter]

I'm interested in taking this on in principal but I have a few deadlines coming up and thus fairly swamped with other work at the moment. 3-4 weeks is a time frame where I can commit to getting something done. If anybody else wants to give it a try I'd be happy to get them jump-started with what I have.

[kwgoodman]

That is great news!

[Midnighter]

I've started to put something together over at this repository but I'm finding it really hard to set up properly only using AppVeyor. Does anyone have a windows machine available for local testing?

[kwgoodman]

Is appveyor giving any feedback about what is not working?

[Midnighter]

It does, it's just a slow workflow since I have to push something to the repository then wait for AppVeyor in order to find out if it works or not. There are just a few things that'd be quicker to figure out locally.

[Midnighter]

You can follow the current status on AppVeyor here. Not even sure that it install Miniconda correctly at this point.

[kwgoodman]

Sounds like a painful process.

[Midnighter]

Good news, the test project finally succeeds! I can now add AppVeyor CI testing to bottleneck. Which branch should I start with, master, c_rewrite, or a new one based off of one or the other?

[kwgoodman]

That's great!

Good timing. As of Friday the c_rewrite branch builds on Windows. How about if you branch off of c_rewrite?

After I remove cython we can then tweak the tests.

[kwgoodman]

I'm going to start removing cython from bottleneck. So you might want to wait for that to be done before adding appveyor. I'll let you know when I'm done.

[kwgoodman]

OK, I merged the c_rewrite branch into master. So you can work off of master now.

BTW I made some changes to travis: I removed cython and I removed the ci directory (ci/travis --> tools/travis).

[kwgoodman]

Bottleneck 1.2 is pretty much done. The only big ticket item is appveyor. Do you know when you'll have time to work on it?

[Midnighter]

I'm at a conference until Sunday. I feel confident I can set up the AppVeyor testing in the week after that. Maybe already on Monday.

[kwgoodman]

Perfect.

[Midnighter]

So basic building is set up. I did not include testing yet. You can see the results here. Lots of warnings about type conversions and some 64bit versions don't build. Do you want me to issue a pull request with this?

[kwgoodman]

Great work! I don't mind waiting for the unit tests. Probably better to have the tests running when making changes for window compat.

[Midnighter]

So do you want a pull request so that you can play with it here or do you want to work off of my fork?

[kwgoodman]

I'll start using it after the unit tests are in place. If I start making changes without the unit tests I may break things without knowing it. But starting a PR is always good, any more commits you make to your branch will be automatically included in the PR. I wouldn't merge until you said it was done.

[Midnighter]

Alright, added testing and those versions that build successfully also seem to be running the tests alright. It's all in #147.

[kwgoodman]

@Midnighter I merged your appveyor PR into master. Appveyor runs are here: https://ci.appveyor.com/project/kwgoodman/bottleneck/history

I see three issues:

• compiler warnings (casting)
• link errors on 64 bit
• why does appveyor make a call to pypi.python.org?

The link errors are, of course, the most important. The first link error:

reduce.obj : error LNK2019: unresolved external symbol __imp_PyErr_Format
referenced in function _import_array

I don't know how to fix the link errors. @cgohlke, do you know the fix for the link errors?

[Midnighter]

why does appveyor make a call to pypi.python.org?

It seems to check the information about pip from https://pypi.python.org/pypi/pip/json, discovers that the cached information is old and retrieves them again. Why? I'm not sure. Might be a conda thing. Doesn't seem to be doing any harm, though.

[cgohlke]

link errors on 64 bit

looks like you are using a 32-bit Python with a 64 bit compiler.

[kwgoodman]

@Midnighter do you know how to fix it? On the log for 64-bit py27 I see (note the 32): creating build\lib.win32-2.7. I'm not even sure which of our appveyor files is responsible.

[Midnighter]

Unfortunately, I have no clue. I have a hunch that it's the MSVC used and therefore clues should be in tools/appveyor/windows_sdk.cmd but from what I see the correct logic is applied and it should be a build for 64bit.

Puzzlingly, for Python 3.5 64bit it also uses the 32bit library and yet the build and tests succeed. I wonder if I should toss this onto StackOverflow or some other forum.

[kwgoodman]

@cgohlke do you have any guidance?

[cgohlke]

It looks like all conda/Python environments created are 32-bit. Maybe I am missing something, but I don't see PYTHON_ARCH used anywhere in conda_wrapper.py

[Midnighter]

It's a left over from dealing with CONDA_FORCE_32BIT which I didn't think was needed. On 64bit it should be undefined anyway and since we're dealing with 32bit on a native 32bit system that shouldn't be an issue either. However, I will play with setting that variable and see if it helps.

[Midnighter]

According to my attempts this has nothing to do with CONDA_FORCE_32BIT but I figured out that there are 32bit and 64bit installs of Miniconda on any system. So it was just a matter of specifying the right path. Pull request is open #148.

[kwgoodman]

In previous releases Cython took care of most of Windows compatibility. Now, without Cython, we are on our own. So thank you @Midnighter for adding AppVeyor. It also gives us our first 32-bit test.

[kwgoodman]

@Midnighter after your AppVeyor fix I am seeing travis failures on osx. I don't see how it could be related. Do you use osx? If so, do the unit test pass on your computer? See #150

[Midnighter]

Sorry, I'm a Linux guy. More on that in the other issue discussion.