search

conda-forge / numba-feedstock

A conda-smithy repository for numba.
BSD 3-Clause "New" or "Revised" License
0 stars 27 forks source link

numba v0.39.0 #13

Closed regro-cf-autotick-bot closed 6 years ago

regro-cf-autotick-bot commented 6 years ago

It is very likely that the current package version for this feedstock is out of date. Notes and instructions for merging this PR:

  1. Please check that the dependencies have not changed.
  2. Please merge the PR only after the tests have passed.
  3. Feel free to push to the bot's branch to update this PR if needed.
  4. The bot will almost always only open one PR per version.

This PR was created by the cf-regro-autotick-bot. The cf-regro-autotick-bot is a service to automatically track the dependency graph, migrate packages, and propose package version updates for conda-forge. If you would like a local version of this bot, you might consider using rever. Rever is a tool for automating software releases and forms the backbone of the bot's conda-forge PRing capability. Rever is both conda (conda install -c conda-forge rever) and pip (pip install re-ver) installable. Finally, feel free to drop us a line if there are any issues!

Here is a list of all the pending dependencies (and their versions) for this repo. Please double check all dependencies before merging.

Name Upstream Version Current Version
llvmlite 0.24.0 Anaconda-Server Badge
conda-forge-linter commented 6 years ago

Hi! This is the friendly automated conda-forge-linting service.

I just wanted to let you know that I linted all conda-recipes in your PR (recipe) and found it was in an excellent condition.

conda-forge-linter commented 6 years ago

Hi! This is the friendly automated conda-forge-linting service.

I just wanted to let you know that I linted all conda-recipes in your PR (recipe) and found it was in an excellent condition.

I do have some suggestions for making it better though...

For recipe:

conda-forge-linter commented 6 years ago

Hi! This is the friendly automated conda-forge-linting service.

I just wanted to let you know that I linted all conda-recipes in your PR (recipe) and found it was in an excellent condition.

jakirkham commented 6 years ago

Hmm...looks like we are getting some test failures. @stuartarchibald, do you have any ideas?

jakirkham commented 6 years ago

@conda-forge-admin, please re-render.

conda-forge-linter commented 6 years ago

Hi! This is the friendly automated conda-forge-webservice.

I tried to re-render for you, but it looks like there was nothing to do.

stuartarchibald commented 6 years ago

@jakirkham here's some initial guesses :)

The jinja2 issue was noted after the 0.39 release and mainline is patched against it here https://github.com/numba/numba/pull/3156. Essentially, there is at least one test that needs jinja2 to pass so updating the recipe to make sure it is there at test time should just solve it (as noted by @mariusvniekerk in https://github.com/conda-forge/numba-feedstock/pull/13/commits/fcbfdeef6f0a9340baf4a2e6d9a744e9d4e6359a).

The failure in np.linalg.cond testing may be down to BLAS impl differences/NumPy+compiler combination differences. I seem to recall that armv7l fails with similar but x86{,_64} and POWER8 do not. The armv7l build would be vanilla NumPy + system compiler which may be similar to the situation here? The test is checking that np.linalg.cond([[1e308, 0], [0, 0.1]]) raises an overflow warning that should come from computing the singular values and dividing large by small, i.e. np.true_divide(1e308, 1e-1).

As to the rest of the failures...

It's probably worth adding numba -s somewhere into the testing scripts so that diagnostic output about the machine on which Numba is running is dumped into the log. Numba's own recipes do it here: https://github.com/numba/numba/blob/master/buildscripts/condarecipe.local/run_test.sh#L29-L30

stuartarchibald commented 6 years ago

The np.linalg.cond fail is valid, it's due to a NumPy change here: https://github.com/numpy/numpy/commit/76f0eafbaba73815710612bddf4bf6eef682965d, the behaviour has changed to be more correct. The armv7l fail for this was because I've been using NumPy master for ARM dev for NumPy and Numba, hence it includes the patch above, whereas the Numba released versions are using NumPy 1.14 as latest. I've added https://github.com/numba/numba/issues/3215 as a reminder to patch this for Numba version 0.40. Given the failure is seemingly harmless, patching in a test skip should be fine (or I can just go write the patch for 0.40 and it can be backported here if that's helpful?).

mbargull commented 6 years ago

Looks like this works much more smoothly with numba 0.40.0: Closing in favor of gh-15.