-
Notifications
You must be signed in to change notification settings - Fork 24
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
Matrix is sometimes incorrect #49
Comments
Weird. Now I can't reproduce it. 😕 |
Closing until we are able to reproduce. Could easily have been a conda-build transient issue. |
Struggling to reproduce this on my local machine, but have the same problem on heroku. I'm going to track it down and will update here. |
Using the following script on the cartopy-feedstock:
I see an empty build matrix on Heroku:
vs. on OSX (where I don't see the problem):
The diff of these two isn't particularly illuminating:
|
When I look at just the index keys, overwhelmingly on OSX the keys include the channel. On Heroku, none appear to have the channel... |
On Heroku the resolve believes that the constraints |
Solved by adding the conda-forge channel. That is embarrassing.... suggests that conda-smithy should be run with the |
Maybe not solved... hmmm, this is a tricky one to track down. |
No worries, @pelson. Agreed this is a slippery one. Maybe this doesn't help, but it might be worth keeping in mind. All the dependencies appear the same. So maybe it is something subtler like the order they were installed/upgraded. Perhaps there is a caching issue like ( conda/conda#3319 )? We've also seen channels misidentified before ( conda-forge/openssl-feedstock#3 (comment) ). Maybe something there too. |
OK, blowing out the package cache on my OSX and I was able to reproduce. I have now tracked this down to:
If I keep the |
The difference being that cartopy is pulling in msinttypes for windows only - it is this that is causing the empty resolve. Assumption is therefore that the index is no longer being "fudged" in the right way to get the windows index and the package dependencies cannot be resolved - hence, an empty build matrix. |
Confirmed. The bug is with fudge_subdir not fudging the right things for the index to be for the correct platform. |
Was a conda-smithy issue (whereby we were trying to control the platform). conda-forge/conda-smithy#274 fixes. |
I'm starting to suspect this is related or is the same as other issues. Like issue ( conda-forge/conda-smithy#257 ). |
Are we really sure that all variant of this bug are fixed? How come is it we are getting odd re-renderings still? |
Turned out to be because we had a different set of channels available to us (conda-forge vs no conda-forge). |
I'm running into this issue with 0.13.0 and 0.13.1, but not with 0.12.0. It sometimes fails to get all of the items in a matrix that I would expect. An example would be trying to run
conda smithy rerender
on this feedstock at that commit. It only provides Python with 3.5 and not any earlier versions. Though there is no version constraint or skip provided. I have provided all of the version info below to help narrow this down.The text was updated successfully, but these errors were encountered: