TST: fix travis, use conda #47
Conversation
|
The first commit allows the travis build to get to the AFAIU sasmodels needs to be in a parallel directory to sasview. At the start of the |
|
Further investigation reveals that the |
|
(1) numpy version on travis is newer, so need to fix sas/sascalc/calculator/slit_length_calculator.py so that (2) matplotlib is complaining that pyparsing isn't > 1.5.6. Since miniconda is installing matplotlib it seems that this is a problem with miniconda. (3) There are numerous errors with the data loader that don't occur when I run tests from my local install. These may have been fixed. Make sure your fork is up to date with master. (4) Long term, the travis build system should mirror the ESS linux build system for the linux distribution. Put the scripts to set up the Ubuntu build environment inside the repo itself and have travis run that. (5) There is a separate question of whether the test system should use the latest anaconda version (since that's what new developers will see) or if it should be fixed at a particular version. (6) travis isn't recognizing that the doc build is partly failing because bumps isn't installed where it is expecting it (part of the docs are pulled in from the bumps repo). Line 2277 of the travis output. (7) building docs requires importing gui modules which requires wx which is not part of the miniconda package list. Line 2370 of the travis output. (8) similarly docs requires py2exe (line 3533). Fixing this will require restructuring the tree to separate the build system files from the packaging. I did this in my own fork before the last code camp: https://github.com/pkienzle/sasview; it is probably worth doing after the 4.1 release. I created ticket #887 for this. |
|
(1) I'll leave this up to someone who knows that section.
(2) installing matplotlib via pip solves this.
(3) I rebased on master, issue is apparently still there.
(6) installing bumps repo solves this
(7) now installing wxpython via conda
(8) I have no idea about this. The sphinx setup looks complicated. I am
trying disabling building of the docs in the travis_build.sh script.
So the errors resulting from (1) and (3) are stalling us here.
…On 21 March 2017 at 03:26, Paul Kienzle ***@***.***> wrote:
(1) numpy version on travis is newer, so need to fix
sas/sascalc/calculator/slit_length_calculator.py so that ind = 0 rather
than ind = 0.0.
(2) matplotlib is complaining that pyparsing isn't > 1.5.6. Since
miniconda is installing matplotlib it seems that this is a problem with
miniconda.
(3) There are numerous errors with the data loader that don't occur when I
run tests from my local install. These may have been fixed. Make sure your
fork is up to date with master.
(4) Long term, the travis build system should mirror the ESS linux build
system for the linux distribution. Put the scripts to set up the Ubuntu
build environment inside the repo itself and have travis run that.
(5) There is a separate question of whether the test system should use the
latest anaconda version (since that's what new developers will see) or if
it should be fixed at a particular version.
(6) travis isn't recognizing that the doc build is partly failing because
bumps isn't installed where it is expecting it (part of the docs are pulled
in from the bumps repo). Line 2277 of the travis output.
(7) building docs requires importing gui modules which requires wx which
is not part of the miniconda package list. Line 2370 of the travis output.
(8) similarly docs requires py2exe (line 3533). Fixing this will require
restructuring the tree to separate the build system files from the
packaging. I did this in my own fork before the last code camp:
https://github.com/pkienzle/sasview; it is probably worth doing after the
4.1 release. I created ticket #887 for this.
—
You are receiving this because you authored the thread.
Reply to this email directly, view it on GitHub
<#47 (comment)>, or mute
the thread
<https://github.com/notifications/unsubscribe-auth/AAq51svnkRu2m7li2e0_8Oe47Z7H50n-ks5rnqijgaJpZM4Mh8Gh>
.
--
_____________________________________
Dr. Andrew Nelson
_____________________________________
|
|
Re: (1), the author of the code is no longer on the project; no one is particularly familiar with it. Re: (3), please turn on the actual test step so I can see how it is failing again Re: (8), building the docs isn't causing travis to complain; leave it in for now so that we can still see the build output. I'm not sure how miniconda is interacting with system packages. I would be tempted to go with apt packages all the way (except pyopencl) since that's how I'm running on my version of ubuntu. |
|
(3) failing readers may be due to floating point indexing into numpy arrays in the IGOR reader. This is moved to ticket #888, which won't be addressed until 4.2. I'm going to close this merge request for now, and you can reopen it when #888 is addressed. |
|
Perhaps the numpy version should be pinned in requirements.txt then?
…On 21 March 2017 at 23:29, Paul Kienzle ***@***.***> wrote:
Closed #47 <#47>.
—
You are receiving this because you authored the thread.
Reply to this email directly, view it on GitHub
<#47 (comment)>, or mute
the thread
<https://github.com/notifications/unsubscribe-auth/AAq51qp8dnW1bbGi-cCVo49Ba5kWXSjHks5rn8K_gaJpZM4Mh8Gh>
.
--
_____________________________________
Dr. Andrew Nelson
_____________________________________
|
Currently the travis build is failing. This PR will fix that.