[Python-modules-team] Bug#936745: reducing matplotlib2 build-depends.
peter green
plugwash at p10link.net
Tue Nov 12 14:13:51 GMT 2019
matplotlib2 seems to be an important node in the python2 removal/reduction problem (and the qt4 removal problem). I have noticed there are a substantial number of python module packages that it build-depends on but does not depend on.
python-backports.functools-lru-cache
python-cairocffi
python-colorspacious
python-cycler
python-functools32
python-gi
python-ipywidgets
python-kiwisolver
python-mock
python-nose
python-numpy
python-numpy-dbg
python-numpydoc
python-pandas
python-pil
python-pkg-resources
python-pyqt5
python-pytest
python-qt4
python-scipy
python-setuptools
python-six
python-sphinx
python-sphinx-gallery
python-subprocess32
python-tk
python-tk-dbg
python-tornado
python-wxgtk3.0
python-xlwt
There is also the slightly-strange case of kiwisolver where there is no binary dependency on python-kiwisolver but there is one on python-kiwisolver-dbg.
Some of these dependencies are starting to cause problems. For example
ipywidgets needs to drop it's python2 packages because jupyter-notebook has already done so, but it can't because of the build-dependency from matplotlib2
the qt developers want to get rid of qt4, but can't because of the build-dependency from matplotlib2.
the pandas maintainers have packaged a new python3 only version, which leaves matplotlib2 build-depending on a cruft package.
I am guessing that many of these are to get testsuite coverage for optional features and are not strictly needed for the build, while testing stuff is nice I don't think it's vital for software that is on it's way out. I tried removing all of the aforementioned packages except python-setuptools and python-kiwisolver from the build-depends (and I uninstalled all python 2 packages from the chroot where I was doing my test builds before I installed the remaining build-depends).
I had to add the following packages back in to get a succesful build.
python-numpy
python-numpy-dbg
python-sphinx (needed for documentation build)
python-backports.functools-lru-cache (needed for documentation build)
python-cycler (needed for documentation build)
python-numpydoc (needed for documentation build)
python-sphinx-gallery (needed for documentation build)
python-colorspacious (needed for documentation build)
python-mock (needed for documentation build)
I also had to add a build-dependency on python-ipkernel which is needed by the documentation build and was previously pulled in indirectly.
That left the following list of changes to build-dependencies.
- python-cairocffi [!ia64],
- python-functools32,
- python-gi,
- python-ipywidgets,
+ python-ipykernel,
- python-nose,
- python-pandas [!hppa !m68k !powerpcspe !sparc64 !sh4 !x32],
- python-pil,
- python-pkg-resources,
- python-pytest,
- python-qt4,
- python-pyqt5 [!hurd-i386],
- python-scipy,
- python-six (>= 1.4),
- python-subprocess32,
- python-tk (>= 2.5.2-1.1),
- python-tk-dbg (>= 2.5.2-1.1),
- python-tornado,
- python-wxgtk3.0,
- python-xlwt,
Unfortunately the matplotlib2 build is not reproducible, so while I was able to determine there were no significant changes to file lists (there were some id changes to id's in the documenation) I was not able to reasonably check for changes to file content.
What do others with more experience think? Should these build-dependencies be removed?
While working on this I also ran into an issue with the clean target not cleaning up properly which I fixed as it was getting in the way of my testing.
debdiff is attached.
-------------- next part --------------
diff -Nru matplotlib2-2.2.4/debian/changelog matplotlib2-2.2.4/debian/changelog
--- matplotlib2-2.2.4/debian/changelog 2019-09-17 03:44:50.000000000 +0000
+++ matplotlib2-2.2.4/debian/changelog 2019-11-12 10:59:29.000000000 +0000
@@ -1,3 +1,13 @@
+matplotlib2 (2.2.4-2.1) UNRELEASED; urgency=medium
+
+ * experimental local build
+ * reduce build-dependencies to (mostly) match binary ones.
+ * add build-depends on python-ipykernel, the documentation build
+ needs it (previously it was pulled in indirectly)
+ * fix clean target.
+
+ -- Peter Michael Green <plugwash at debian.org> Tue, 12 Nov 2019 10:59:29 +0000
+
matplotlib2 (2.2.4-2) unstable; urgency=medium
* debian/control
diff -Nru matplotlib2-2.2.4/debian/control matplotlib2-2.2.4/debian/control
--- matplotlib2-2.2.4/debian/control 2019-09-17 03:44:50.000000000 +0000
+++ matplotlib2-2.2.4/debian/control 2019-11-12 10:59:29.000000000 +0000
@@ -19,36 +19,18 @@
python-cycler (>= 0.10.0),
python-dateutil,
python-colorspacious,
- python-cairocffi [!ia64],
- python-functools32,
- python-gi,
- python-ipywidgets,
+ python-ipykernel,
python-kiwisolver,
python-kiwisolver-dbg,
python-mock,
- python-nose,
python-numpy,
python-numpy-dbg,
python-numpydoc,
- python-pandas [!hppa !m68k !powerpcspe !sparc64 !sh4 !x32],
- python-pil,
- python-pkg-resources,
python-pyparsing (>= 1.5.6),
- python-pytest,
- python-qt4,
- python-pyqt5 [!hurd-i386],
- python-scipy,
python-setuptools,
- python-six (>= 1.4),
python-sphinx (>= 1.0.7+dfsg),
python-sphinx-gallery,
- python-subprocess32,
- python-tk (>= 2.5.2-1.1),
- python-tk-dbg (>= 2.5.2-1.1),
- python-tornado,
python-tz,
- python-wxgtk3.0,
- python-xlwt,
tcl8.6-dev,
texlive-fonts-recommended,
texlive-latex-extra,
diff -Nru matplotlib2-2.2.4/debian/rules matplotlib2-2.2.4/debian/rules
--- matplotlib2-2.2.4/debian/rules 2019-09-17 03:44:50.000000000 +0000
+++ matplotlib2-2.2.4/debian/rules 2019-11-12 10:59:29.000000000 +0000
@@ -93,6 +93,22 @@
# remove embedded CXX to prevent it to be picked up
rm -rf extern/CXX/
+ #more cleanup
+ rm -rf .eggs
+ find examples -name '*.png' -delete
+ find examples -name '*.pdf' -delete
+ find examples -name '*.svgz' -delete
+ rm -f examples/misc/*.svg
+ find . -name '*.zip' -delete
+ find . -name '*.md5' -delete
+ find . -name '*.pickle' -delete
+ find doc -name '*.ipynb' -delete
+ rm -rf doc/tutorials
+ rm -rf doc/gallery
+ rm -rf doc/build
+ rm -f doc/fontList.json
+ rm -f examples/misc/*.eps
+
# install
install-arch: build-arch install-arch-stamp
dh_testdir
More information about the Python-modules-team
mailing list