-
-
Notifications
You must be signed in to change notification settings - Fork 7.9k
Matplotlib 1.4.2 fails to build in Python 3 (TypeError: unorderable types: str() < int()) #3889
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
Comments
Do you have libpng installed? Something funny is coming out of |
I do, this is what I get:
|
But during the build it returns |
I wonder why this does not blow up on travis... This is yet more issues with unicode vs bytes. |
I think I found it. We install our own libpng package, but for some reason
So that's complete garbage. It's probably related to the fact that we use pkgconf: http://rabbit.dereferenced.org/~nenolod/distfiles/pkgconf-0.9.6.tar.bz2, instead of the usual pkg-config. In any case, let's see if I can fix this. |
@tacaswell I found it!! This patch fixes it: hashdist/hashstack@0aa2605, what happened was that we were putting our own wrapper on top of So this issue is fixed, it's unrelated to matplotlib. But I still think the error message might be more descriptive --- maybe you can check that the answer makes sense? |
Thanks for your help, this would take me a lot longer to figure out without it. |
The explanation makes sense. I would argue that the real problem is in distutils. We are using |
Ah I see. Yes, indeed. Quite frankly, I just use cmake for my projects and I handle Python installation from cmake. Then I don't have any of these issues. |
Yes, but then you have all manner of cmake issues 😉 Your projects won't (might not?) play nice with other people's setup.py/pip/requirements.txt/venvs. I had an absolute bear of a time trying to get peoples cmake built python packages (ITK) to play nice with virtual environments. Of course the solution to that is hashdist/conda (which at least in the case of conda were created to deal with this pain). |
Exactly, the same for Hashdist. I think it is possible to make cmake play nice with pip, e.g. here we test that at least |
are we sure this is fixed? I'm getting this on 1.4.3:
|
Does the same happen with 1.4.3? That very unhelpful error is saying that you do not have freetype-dev or Tom On Thu, Apr 16, 2015, 17:51 John Anderson [email protected] wrote:
|
@tacaswell It is it 1.4.3, here is more info:
|
hahaha, I found out why its failing...
So I have GREP_OPTIONS in my bashrc and that is deprecated, and I guess the way you are executing the command its catching those warnings as well. |
Installation worked after I did:
My grep looked like this before:
|
That is new and interesting. I don't think there is much we can do about Sorry for the version confusion, I read 1.4.2 not what you actually wrote. On Thu, Apr 16, 2015, 18:13 John Anderson [email protected] wrote:
|
@tacaswell Could you send stderr to dev null like:
and only grab stdout? |
Having same issue - Python3.5.0a3 (dev) / Mac OSX
|
@ben-albrecht Can you make sure that you have freetype installed. Alternatively try the master branch we recently fixed a bug in the version comparison only happening when freetype is not found |
@jenshnielsen Thanks for the suggestion. The latest version from master branch gives a more helpful error message
I Installed Freetype via homebrew and pip installed matplotlib from via |
I'm seeing a similar failure in debian unstable:
and that line seems to refer indeed to freetype, but running -ftversion i get:
any idea what might be going on? |
do you have |
yes pkg-config is installed (at version 0.29-4 if that's useful to know) |
@sandrotosi can you open a new issue for this? |
i rebuilt the packages just to be sure (i had a suspect, reading this bug, it might have been due to a library (fakeroot) that was generating symbols errors and thus messing up with the versions gathering) and it was a transient error, which is now over - sorry for the noise |
@sandrotosi I am facing same issue with python 3.4.3 and pip3 install matplotlib==1.4.3:
|
Ok, guys as soon as on one system the error did not appear and on other did I compared libs and found issue, there were missing:
Is Python really that stupid or it is due to how matplotlib is managed? If I miss library I screem I am missing that library, so user can install it.... this is detective work... |
The error you are getting is in the logic where we are trying to tell you that there are missing libraries. You are trying to build no longer supported versions of Matplotlib on an old version of python. I believe we have fixed all of these issues in the current versions of Matplotlib with up-to-date python. |
I know, this is some software we received from vendor to install in our environment, not much to do. Apps are developed and used without update for years... When I am in Java and miss some library it tells me: you are missing this library FULL NAME of class, so... But ok, I know this is old thread... thx for answer anyway. |
As part of Hashdist:
The issues #2087 and #3732 seem related.
It works in Python 2.7. The dependencies are exactly the same for 2.7 and 3.4. Here is the build script:
And I used the 1.4.2 tarball: https://downloads.sourceforge.net/project/matplotlib/matplotlib/matplotlib-1.4.2/matplotlib-1.4.2.tar.gz
And I use linux (RHEL6), though we build all the dependencies ourselves.
The text was updated successfully, but these errors were encountered: