From tom.augspurger88 at gmail.com Sat Jun 11 10:55:15 2016 From: tom.augspurger88 at gmail.com (Tom Augspurger) Date: Sat, 11 Jun 2016 07:55:15 -0700 (PDT) Subject: [Wheel-builders] Double check my pandas wheel repo? Message-ID: <52p2e77n9ziqqoke41u1xxrnv-0@mailer.nylas.com> Hi, Does anyone have a few minutes to check the wheel-builder I put together for pandas here[1]? The build script here[2] is pretty similar to the example, but without the test step since that was timing out on Travis. I just worry that since pandas is somewhat complex, and since I'm not a C programmer, I could be missing something. A wheel built in a local docker container only had a couple test failures we had already seen, but I wanted to double-check with the experts before putting this on PyPI :) Thanks, Tom [1]: https://github.com/TomAugspurger/pandas-manylinux [2]: https://github.com/TomAugspurger/pandas-manylinux/blob/master/travis /build-wheels.sh -------------- next part -------------- An HTML attachment was scrubbed... URL: From matthew.brett at gmail.com Sat Jun 11 15:53:21 2016 From: matthew.brett at gmail.com (Matthew Brett) Date: Sat, 11 Jun 2016 12:53:21 -0700 Subject: [Wheel-builders] Double check my pandas wheel repo? In-Reply-To: <52p2e77n9ziqqoke41u1xxrnv-0@mailer.nylas.com> References: <52p2e77n9ziqqoke41u1xxrnv-0@mailer.nylas.com> Message-ID: Hi, On Sat, Jun 11, 2016 at 7:55 AM, Tom Augspurger wrote: > Hi, > > Does anyone have a few minutes to check the wheel-builder I put together for > pandas here[1]? > The build script here[2] is pretty similar to the example, but without the > test step since that was timing out on Travis. > I just worry that since pandas is somewhat complex, and since I'm not a C > programmer, I could be missing something. > A wheel built in a local docker container only had a couple test failures we > had already seen, but I wanted to double-check with the experts before > putting this on PyPI :) I make some suggestions over at https://github.com/TomAugspurger/pandas-manylinux/pull/3 You probably know, but iterating through all the directories in /opts/python will pick up all Pythons, which may in the future include things like pypy - maybe this doesn't matter but I thought I'd mention it. Cheers, Matthew From urschrei at gmail.com Sun Jun 12 15:07:19 2016 From: urschrei at gmail.com (=?utf-8?Q?Stephan_H=C3=BCgel?=) Date: Sun, 12 Jun 2016 21:07:19 +0200 Subject: [Wheel-builders] Manylinux Docker image isn't including shared library for a Cython extension Message-ID: <480EFA5C-516C-40F2-A8B8-67FB53D73C9B@gmail.com> I'm attempting to build a Python package using the Manylinux docker image. The package includes a Cython extension which requires a .so file, which I'm including in the extension directory. I've built the .so with rpath support, and I'm passing $ORIGIN as an extra link argument for Linux builds (see setup.py here: https://github.com/urschrei/convertbng/blob/master/setup.py). This works correctly on everything except Manylinux; I can build the package using bdist_wheel on a Linux host, and the .so is correctly copied into the wheel, all tests pass, and installing the built wheel on another host also works. Building on an OSX image on Travis also works correctly. However, when I build it inside the Docker image, a number of files aren't copied into the wheel, including the .so required by the Cython extension, but the wheel build completes anyway. I suspect I'm overlooking some subtlety here, but I'm not sure what's going on. Here's my latest Travis build log using the Manylinux image: https://travis-ci.org/urschrei/convertbng/jobs/137096797#L705 (the successful OSX build is #260.2) Could anyone help me figure out what's going on? -- steph -------------- next part -------------- An HTML attachment was scrubbed... URL: From olivier.grisel at ensta.org Mon Jun 13 03:38:45 2016 From: olivier.grisel at ensta.org (Olivier Grisel) Date: Mon, 13 Jun 2016 09:38:45 +0200 Subject: [Wheel-builders] Manylinux Docker image isn't including shared library for a Cython extension In-Reply-To: <480EFA5C-516C-40F2-A8B8-67FB53D73C9B@gmail.com> References: <480EFA5C-516C-40F2-A8B8-67FB53D73C9B@gmail.com> Message-ID: liblonlat_bng.so should also be compiled within the manylinux1 image otherwise it will depend on a version of glibc that is too recent to meet the manylinux1 contract. This seems to be the problem reported by the output of the "auditwheel show" command: https://travis-ci.org/urschrei/convertbng/jobs/137096797#L755 -- Olivier