New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Appveyor overhaul #6520
Appveyor overhaul #6520
Conversation
Looks good, Any chance we can drop the patched bdist_whell now? |
Yes, that can also be gone now. Will add it here in this PR |
e124896
to
8b0198c
Compare
Meh: [matplotlib-1.5.0+2326.gf349629-cp27-cp27m-win_amd64.whl]
-> depends on |
361164b
to
22be5ae
Compare
ea676d3
to
22c7f7b
Compare
Ok, it builds and the tests pass but the coverage is going down :-( The problem with non-static dependencies is fixed as far as I can see and the tests now include a test for that. -> I would love to get a review, I don't have any more changes staged for this. |
Don't worry about the coverage. It's comming from the travis builds and have been going up and down for that file all the time in various builds |
129a9e8
to
a4b2e1a
Compare
Ok, after installing inkscape and imagemagick the tests error out:
The test runtime of almost exactly 1h seem to suggest that the tests are stuck somewhere and run into a timeout :-( -> I'm removing the choko installs again... |
Ok, wasn't the choco lines, next try pillow... Please don't let it be miktex... |
Re possible MiKTeX timeouts: check that MiKTeX is configured to "install missing packages on-the-fly", not "ask me first" (the default; opens a modal dialog). |
@cgohlke: Thank you, the miktex "ask" setting seems to have been the problem!
-> It seems it still misses most of the tests due to missing tools :-( probably the image conversation util to convert svg/... to png |
The images are converted using inkscape but I guess choko installing inkscape does not put it on path so it can be found by nose |
Current status: A failure on py3.4: https://ci.appveyor.com/project/mdboom/matplotlib/build/1.0.1712
Next stop: trying to put inkscape into PATH thanks to the suggestion from @jenshnielsen |
@JanSchulz I think that one is a random fluke. We are seeing it on Travis too |
This is mainly because the additional tests (which are otherwise skipped) increase the runtime of one test run (we have 4) from 13min to 30mins and thereby increasing one complete Appveyor run from 40min to 2 hours. With this change, we only have 30+3*13min. The py27 test was chosen because up it made the most problems during this Appveyor/Windows fixing round...
e25b820
to
e7c9b56
Compare
I've disabled the conda-build and removed the debugging output (PR for both is in #6682), so hopefully this succeeds now and can be merged... |
The tests are now all successful. Who should be the one to give the final review and check? |
@@ -11,4 +11,4 @@ index 8af8b6d..4e4f9d2 100644 | |||
+ f.write(__version__) | |||
|
|||
# These are the packages in the order we want to display them. This | |||
# list may contain strings to create section headers for the display. | |||
# list may contain strings to create section headers for the display. |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Don't think this needs to be changed?
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
yep, will change
Again, don't use `load_setuptools()` but the __conda_version__.txt file to get the final version of the package from the git commit id.
e7c9b56
to
32fb993
Compare
That seems to be the worst platform, so should trigger the most issues :-)
Do we version pin something to make sure we are using a fixed version of the wheels build code? |
@tacaswell No, they are basically build against the same code which is in conda-forge at this time. |
I am a bit concerned about removing the bug workaround in setup.py which may bite users on windows that want to build wheels which have not updated. I am happy with saying that people must have newer versions of setuptools (I assume that is where it comes from?), but should probably put in a check for that. I would also be happy to be told I am confused about this and it is fine as-is. |
I saw basically two bugreports about the wheel issue: mine and one from cgohlke who also tried compiling mpl. So IMO this is pretty rare: noone in the right mind tries to compile mpl from source to a wheel, everyone just gets it from cgohlke :-) Wheel is from it's own package (at least under conda) but gets probably installed per default (at least conda installs it automatically in each new python env)? |
Fair enough. |
Yay :-) Thanks! |
Thank you for fighting the good fight with appveyor! |
Build times have increased from ~10 min (build <= 1.0.1922) to ~30 min (build >= 1.0.1924) on all four cases (not just one 833f903) after merging this. |
before:
after:
And 64bit and the lone 32 bit (which was instructed to run all image comp tests) have the same runtime and the same number of tests. Should I remove the additional requirements (latex, ffmeg) on all but one build, so only one gets down to 609 skips vs 1200 in the old days? |
On my machine |
librsvg has other issues like not implementing the svg spec correctly (https://bugzilla.gnome.org/show_bug.cgi?id=748565). |
I think @mdboom mentioned that they had fixed that issue but probably not in a release yet |
Mostly: