Re: [gentoo-python] Re: [PATCH 5/8] distutils-r1: disable sub-root merging.

[Mike Gilbert <floppym@gentoo.org>]

On Sat, Sep 14, 2013 at 6:38 PM, Michał Górny <mgorny@gentoo.org> wrote:
> Dnia 2013-09-14, o godz. 18:16:33
> Mike Gilbert <floppym@gentoo.org> napisał(a):
>
>> On Fri, Sep 13, 2013 at 2:58 PM, Michał Górny <mgorny@gentoo.org> wrote:
>> > The sub-roots are no longer necessary since we're installing scripts to
>> > per-impl directories initially.
>> > ---
>> >  gx86/eclass/distutils-r1.eclass | 9 +++------
>> >  1 file changed, 3 insertions(+), 6 deletions(-)
>> >
>> > diff --git a/gx86/eclass/distutils-r1.eclass b/gx86/eclass/distutils-r1.eclass
>> > index 7102c58..5fcb55e 100644
>> > --- a/gx86/eclass/distutils-r1.eclass
>> > +++ b/gx86/eclass/distutils-r1.eclass
>> > @@ -476,9 +476,7 @@ distutils-r1_python_install() {
>> >         addpredict "$(python_get_sitedir)"
>> >         addpredict /usr/lib/portage/pym
>> >
>> > -       local root=${D}/_${EPYTHON}
>> > -       [[ ${DISTUTILS_SINGLE_IMPL} ]] && root=${D}
>> > -       flags+=( --root="${root}" )
>> > +       flags+=( --root="${D}" )
>> >
>> >         if [[ ! ${DISTUTILS_SINGLE_IMPL} ]]; then
>> >                 local PYTHON_SCRIPTDIR
>> > @@ -488,13 +486,12 @@ distutils-r1_python_install() {
>> >
>> >         esetup.py install "${flags[@]}" "${@}"
>> >
>> > -       if [[ -d ${root}$(python_get_sitedir)/tests ]]; then
>> > +       if [[ -d ${D}$(python_get_sitedir)/tests ]]; then
>> >                 die "Package installs 'tests' package, file collisions likely."
>> >         fi
>> >
>> >         if [[ ! ${DISTUTILS_SINGLE_IMPL} ]]; then
>> > -               _distutils-r1_wrap_scripts "${root}"
>> > -               multibuild_merge_root "${root}" "${D}"
>> > +               _distutils-r1_wrap_scripts "${D}"
>> >         fi
>> >  }
>> >
>> > --
>> > 1.8.3.2
>> >
>>
>> I believe this patch causes a race condition which causes
>> java-config's ebuild to fail.
>>
>> running install_data
>> running install_egg_info
>> Writing /tmp/portage/dev-java/java-config-2.2.0/image/usr/lib64/python2.7/site-packages/java_config-
>> 2.2.0-py2.7.egg-info
>> error: /tmp/portage/dev-java/java-config-2.2.0/image//usr/share/java-config-2/config/:
>> File exists
>
> I love when people hack build systems and have no idea about writing
> build systems...
>
> What are you suggesting? Finding all the broken stuff and fixing it, or
> restoring sub-roots?
>

This is a well-used package which will be broken if we apply this
patch. I'm not sure we can reliably identify this kind of problem, and
fixing individual packages would be a bit of work in any case..
Keeping the sub-roots code in place seems like the safer course of
action here.

Also FYI, I looked into java-config's setup.py, and the code has an
unchecked os.mkdir call followed by an open() call. The former will
fail for multiple python implementations even with a serialized
install, and the latter is racy.