Re: [libvirt] [PATCH] Remove python binding

On 11/22/2013 09:59 AM, Daniel P. Berrange wrote: These changes are important... and these... and this (hmm, do we need a ./run counterpart added to libvirt-python to make it easy to import a built but uninstalled python bindings for testing purposes?) The rest of the patch is mechanical 'git rm'; it might have been nice to use git's -O/path/to/file to order the patch to float the interesting parts up top and trim the rest of the email. grepping for python still had hits after your patch; here's my analysis: ChangeLog-old - good to keep; we aren't changing details of historical releases bootstrap.conf - unsure: Does bootstrap.conf still need to worry about finding python-config, or is that a thing of the past? cfg.mk - keep; we still have .py files for docs docs/Makefile.am docs/apibuild.py docs/apps.html.in docs/bindings.html.in docs/errors.html.in docs/hooks.html.in docs/index.py docs/library.xen docs/news.html.in docs/python.html.in docs/sitemap.html.in docs/testsuites.html.in docs/windows.html.in - good to keep all docs stuff for doc generation libvirt.spec.in - good to keep; only mentions the historical %changelog src/esx/esx_vi_generator.py src/hyperv/hyperv_wmi_generator.py src/util/virconf.c src/util/virkeycode-mapgen.py - good to keep all the C code generators We also need to scrub the code base for references to 1.1.5 and instead refer to 1.2.0. Oddly enough, removing these lines will mean that someone with an incremental tree that likes to switch branches between master and any earlier branch (say v1.0.5-maint) will now see git complaining about untracked files. I personally tend to avoid removing lines from .gitignore (if we've ever ignored a file in the past, then branch switching could leave the file around to still be ignored); but I won't complain too much if you make the deletion (I can always re-add the lines in my .git/info/exclude for my own environment). Did configure.ac clean up the setting of $(pythondir)? /me looks some more Wow - we never set $(pythondir) anywhere pre-patch, so it only ever did something if you invoked 'make pythondir=...', and since autobuild wasn't doing it, I think this was dead code. Hmm. Was --without-python disabling _just_ the python bindings, or was it _also_ disabling our attempts to use $(PYTHON) for building things in our docs/ subtree? We may STILL want to keep this setting for controlling whether to attempt to build generated docs (vs. using the pre-generated version shipped in the tarball). Ouch. Deleting this may break uses of $(PYTHON) when building our docs. Meanwhile, this part is probably safe to drop, which goes back to my question on getting rid of the check for python-config in bootstrap.conf. So it looks like pre-patch, --without-python attempted to disable only the bindings, but may have also accidentally disabled the setting of $(PYTHON)... ...maybe all we need is a simple AC_PATH_PROG([PYTHON], [python]) for the sake of still using $(PYTHON) during doc generation? And some of this goes back to whether deleting --without-python from autobuild makes sense. Giant snip :) Ouch. Don't we still need a buildreq on 'python' for doc generation purposes, possibly conditionally based on whether we have patches applied to a downstream rpm that warrant doc regeneration? All of this transferred over to the libvirt-python.git, so good to see it go here. Same question as in autobuild on whether --without-python still matters for suppressing doc regeneration, or whether it was intended only for suppressing bindings. Another giant snip :) Fine. If you can answer the comments I raised above, and if everyone is okay with your libvirt-python.git patches, then I'm okay with this patch. Let's get it polished and in before freeze! I also just kicked off a 'make distcheck' and will report back later if I saw anything fishy that needs fixing. -- Eric Blake eblake redhat com +1-919-301-3266 Libvirt virtualization library http://libvirt.org