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
Upload packages to PyPI directly, for pip 1.5 #2715
Comments
It seems matplotlib is still too big. Uploading through the web interface gives:
Any help you can provide to make this work would be great. |
Does the tarball strip out the testing images? |
So the (obvious) two solutions are strip things out of the tarball to make it smaller, or bump up the file size. Can you tell me what the file size you tried to upload? |
No, the tarball does not strip out the testing images. I'm not sure we want to do that except as a last resort. The source tarball is ~42MB. |
Ok, that's about 2MB over the current limit. The limit is pretty arbitrary so we can bump that up a bit. I'll comment on this ticket once I get time to do that (should be sometime today). |
It's primarily the lib/matplotlib/tests/baseline_images I assume that are making this so big? Is the purpose here so people can test their installation (or development setup)? If people use matplotlib in production settings where they repeatedly deploy it, would it make sense to make a version that doesn't bundle all this in? |
If I wanted to provide packages both with and without the test data, how would you recommend doing that on pypi? Have two separate package entries, i.e. use the existing I don't want to change this for a bugfix release, but rather wait for our next major release (which is scheduled for "pretty soon now"). |
And in the interest of not changing things -- I should probably go and upload to PyPI the last few years worth of releases exactly as they were on Sourceforge -- it's actually pretty typical with matplotlib for users to specify old versions, so I'd like that to work without the |
Installing matplotlib-tests into a namespace package sounds good. I've never used it, but I imagine the syntax in entry_points={
"matplotlib": [
"tests = matplotlib-tests"
]
} I summon @ivanov for more expertise than myself (we discussed this in person). |
I +1 the idea to bump up the file limit (e.g. 50MB should be sufficient). I just recently ran into this problem where only an old version was downloaded. I can't imagine I'm the only one trying to "pip install matplotlib", so it's definitely not good that this doesn't work (especially since I'm using Python 3, where >= 1.2.0 is required). |
@dstufft: What's the status on the file size limit? I think I'd prefer that solution (for now), so as to not change our installation procedure in a significant way in a bug fix release. Is there a PyPI bug tracker entry (or some such) I could follow? |
I didn't find a dedicated issue there, but I reported this problem at https://sourceforge.net/p/pypi/support-requests/342/ and pointed out the file size limitation issue. |
I'm a terrible person and forgot all about this, let me do it right now. |
Can you try to upload to TestPyPI and see if it works there? If it does I'll deploy it out to PyPI itself. |
I get:
How would I further debug this? |
o.O Uhhh, can you try using Twine to upload instead? https://pypi.python.org/pypi/twine/ |
How do I tell twine to go to testpypi?
|
Make sure testpypi is in your ~/.pypirc and use -r testpypi like:
|
FWIW i'm dstufft on freenode if you want to do this in a more immediate response format :) |
|
Ah ha, Hmm Let me go dig through the logs. Did you generate it jsut with |
I believe this should be fixed now on PyPI and TestPyPI |
Success (on TestPyPI). |
On PyPI proper, it seems it still has the file size limit:
Don't know if that's still to be expected. Once that's resolved, I'll upload this, and this bug can be closed... Thanks for all of your help. |
@mdboom just verified the updated limit is in place on PyPI, would you mind trying again? |
Success! Thanks for everyone's help. Closing. |
As of pip 1.5 (released January 1, 2014), pip will no longer install externally hosted files by default.
If a user tries to
pip install matplotlib
it will default to the most recent package hosted on PyPI (1.1.1).Pinning the dependency to install matplotlib 1.3.1 results in a helpful error message
Users can work around this by either specifying to allow external and unverified,
or by installing straight from the package for their respective system (tarball for *nix shown below)
It would be really nice to have this package available for pip. It seems that in the past this was a filesize limit restriction workaround on PyPI, which can now handle bigger packages.
If the filesize is still an issue, perhaps we can work with the PyPI folks to either increase the limit or put matplotlib on a whitelist.
The text was updated successfully, but these errors were encountered: