Matplotlib no longer builds without X11, but there seems to be no reason for it. This seems to have changed since 1.3.0 (i.e. when 1.3.1 was added). The ChangeLog mentions "wrap commands with virtualx.eclass". The package seems to build (and work) if the line VIRTUALX_REQUIRED="always" is removed from the ebuild, though this is of course unlikely to be the correct solution. My guess would have been that it would be specifying something like VIRTUALX_REQUIRED="fltk gtk gtk3 pyside qt4 tk wxwidgets", but it seems VIRTUALX_REQUIRED only supports a single value. Reproducible: Always Steps to Reproduce: # With USE="-fltk -gtk -gtk3 -pyside -qt4 -tk -wxwidgets" emerge matplotlib -a Actual Results: A lot of X11 packages are pulled in. Expected Results: No X11 packages should be pulled in.
matplotlib-1.3.1.ebuild:11:VIRTUALX_REQUIRED="always"
VIRTUALX_REQUIRED="always" may be un-needed.
i use matplotlib with only the cairo USEFLAG on several web servers, and it never required X and worked very well. I highly doubt that upstream changed dep in such a huge way between 1.3.0 and 1.3.1 .. ? Please remove this, portage want to install the whole huge x stack just to update matplotlib ...
And i confirm it works by copying the ebuild to a local overlay, and remove the VIRTUALX_REQUIRED="always"
I was going to file a bug, but found this thread. I use matplotlib on prefix on OS X, and I was surprised to see that matplotlib-1.3.1 depends on xhost directly, while matplotlib-1.3.0 doesn't. I didn't really know why, but I believe that the VIRTUALX_REQUIRED is causing this. Please see some attachments with the dependencies. I'd like to be able to install it without installing all the X related packages as before.
Created attachment 383856 [details, diff] matplotlib-1.3.0-1.3.1.diff
Created attachment 383858 [details] dependencies for 1.3.0 with equery
Created attachment 383860 [details] dependencies for 1.3.1 with equery
Comment on attachment 383858 [details] dependencies for 1.3.0 with equery dev-python/matplotlib-1.3.0: [ 0] dev-python/matplotlib-1.3.0 [ 1] dev-python/numpy-1.8.2 [ 1] dev-python/python-dateutil-2.2 [ 1] dev-python/pytz-2014.4 [ 1] dev-python/six-1.7.3 [ 1] media-fonts/stix-fonts-1.0.0-r1 [ 1] media-libs/freetype-2.5.3-r1 [ 1] media-libs/libpng-1.6.13 [ 1] dev-python/pygtk-2.24.0-r4 [ 1] dev-python/wxpython-2.9.4.1 [ 1] virtual/pkgconfig-0-r1 [ 1] app-text/dvipng-1.14 [ 1] virtual/python-imaging-2 [ 1] dev-python/ipython-2.2.0 [ 1] dev-python/numpydoc-0.4-r1 [ 1] dev-python/xlwt-0.7.5 [ 1] dev-python/sphinx-1.2.2 [ 1] dev-texlive/texlive-latexextra-2013 [ 1] dev-texlive/texlive-fontsrecommended-2013 [ 1] dev-texlive/texlive-latexrecommended-2013 [ 1] media-gfx/graphviz-2.36.0 [ 1] dev-python/nose-1.3.3 [ 1] dev-lang/python-2.7.6-r1 [ 1] dev-lang/python-3.2.5-r3 [ 1] dev-lang/python-3.3.3 [ 1] dev-lang/python-exec-2.0.1-r1 [ 1] dev-python/pyparsing-2.0.2 [ 1] dev-python/pycairo-1.8.10 [ 1] dev-python/pyfltk-1.3.0-r1 [ 1] dev-python/pygobject-3.12.2 [ 1] x11-libs/gtk+-3.12.2 [ 1] virtual/latex-base-1.0 [ 1] app-text/ghostscript-gpl-9.10-r1 [ 1] app-text/poppler-0.26.4 [ 1] dev-python/PyQt4-4.11.1 [ 1] dev-python/pyside-1.2.2
Comment on attachment 383860 [details] dependencies for 1.3.1 with equery dev-python/matplotlib-1.3.1: [ 0] dev-python/matplotlib-1.3.1 [ 1] dev-python/numpy-1.8.2 [ 1] dev-python/python-dateutil-2.2 [ 1] dev-python/pytz-2014.4 [ 1] dev-python/six-1.7.3 [ 1] media-fonts/stix-fonts-1.0.0-r1 [ 1] media-libs/freetype-2.5.3-r1 [ 1] media-libs/libpng-1.6.13 [ 1] dev-python/pygtk-2.24.0-r4 [ 1] dev-python/wxpython-2.9.4.1 [ 1] virtual/pkgconfig-0-r1 [ 1] app-text/dvipng-1.14 [ 1] virtual/python-imaging-2 [ 1] dev-python/ipython-2.2.0 [ 1] dev-python/numpydoc-0.4-r1 [ 1] dev-python/xlwt-0.7.5 [ 1] dev-python/sphinx-1.2.2 [ 1] dev-texlive/texlive-latexextra-2013 [ 1] dev-texlive/texlive-fontsrecommended-2013 [ 1] dev-texlive/texlive-latexrecommended-2013 [ 1] media-gfx/graphviz-2.36.0 [ 1] dev-python/nose-1.3.3 [ 1] dev-lang/python-2.7.6-r1 [ 1] dev-lang/python-3.2.5-r3 [ 1] dev-lang/python-3.3.3 [ 1] dev-lang/python-exec-2.0.1-r1 [ 1] x11-base/xorg-server-1.16.0 [ 1] x11-apps/xhost-1.0.6 [ 1] dev-python/pyparsing-2.0.2 [ 1] dev-python/pycairo-1.8.10 [ 1] dev-python/pyfltk-1.3.0-r1 [ 1] dev-python/pygobject-3.12.2 [ 1] x11-libs/gtk+-3.12.2 [ 1] virtual/latex-base-1.0 [ 1] app-text/ghostscript-gpl-9.10-r1 [ 1] app-text/poppler-0.26.4 [ 1] dev-texlive/texlive-xetex-2013 [ 1] dev-python/pyside-1.2.2 [ 1] dev-python/PyQt4-4.11.1
Sorry for the edits. The first files were garbled with colored output from equery.
I commented out VIRTUALX_REQUIRED="always" and it works on my prefix on OS X as well. Please consider removing that from the dependencies.
(In reply to Guilherme Amadio from comment #12) > I commented out VIRTUALX_REQUIRED="always" and it works on my prefix on OS X > as well. > Please consider removing that from the dependencies. Probably it is using lots of stuff from the host.
mpl is doing lots of testing which fails without X and automagically disables support for some components. That's the reason I added the virtualx usage.
(In reply to Justin Lecher from comment #14) While I understand your rationale for adding that dependency to the ebuild, I think that the expected behavior in this case is to let it work without X, as it always did. I'd rather let mpl disable the components for things I don't have installed than have to install things I will not use. On prefix, it is often the case the the dependencies are keyword masked (which is the case of xhost), and it only makes installing stuff more difficult/annoying. On Linux, X will most likely be installed and the components will not be disabled, but some people might be on Wayland or building mpl on a server and not want X at all. On OS X, X isn't needed either. Therefore, if the package works fine without X, I think we should not make its components hard dependencies. Please consider removing VIRTUALX_REQUIRED="always" from the ebuilds.
Indeed, i have matplotlib installed on all my servers, it's very often a dependancy for some php applications. And there's no X there...
(In reply to Guilherme Amadio from comment #15) > Please consider removing VIRTUALX_REQUIRED="always" from the ebuilds. I will do a careful evaluation and see what I can do.
(In reply to Thomas Capricelli from comment #16) > And there's no X there... it is build time only. Although I understand that it sucks having this heavy buildtime dependency, you are free to remove it later.
17 Sep 2014; Justin Lecher <jlec@gentoo.org> matplotlib-1.4.0.ebuild: Reduce usage of virtualx, #498028
