Bug 14137

Note that this bug *may* reveal a broader issue with python packages.  The package builder should probably auto-detect these dependencies and add them to the rpm package automatically.

I've noticed other problems with python packages.  For example, the python-jinja2 has the following requirements:

# rpm -qp --requires python-jinja2-2.7.3-1.mga5.noarch.rpm
python-babel >= 0.8
python-markupsafe
rpmlib(PartialHardlinkSets) <= 4.0.4-1
rpmlib(PayloadFilesHavePrefix) <= 4.0-1
rpmlib(CompressedFileNames) <= 3.0.4-1
python(abi) = 2.7
pythonegg(2)(markupsafe)
rpmlib(PayloadIsXz) <= 5.2-1

Note that the markupsafe module is added both manually (I guess) as âpython-markupsafeâ, and automatically as âpythonegg(2)(markupsafe)â.  Yet the markupsafe package provides the following:

# rpm -qp --provides python-markupsafe-0.23-2.mga5.x86_64.rpm
python-markupsafe = 0.23-2.mga5
python-markupsafe(x86-64) = 0.23-2.mga5

So right now the python-jinja2 package has a missing dependency.  If it is true that the automatic dependency finder is broken for python modules, this may not be a junior job and it should probably be fixed before the final release.  Honestly I have no idea if this is true or not.

There are many more problems.  (For example, as far as I can tell, the rpmdrake package dependencies are broken too.)  As soon as the first beta is released, I will reinstall the system on my computer and I will report python dependency issues I run into.

The python automatically generated dependencies were in fact broken in Cauldron, but if I understand correctly, they have been fixed.  Is this bug still valid?

CC: (none) => makowski.mageia, thierry.vignaud

Yes, as far as I can tell, the bug is still valid.  The current python-flask package (0.10.1-6.mga5, build on Sep 27) basically has no module dependency at all:

$ rpm -q --requires python-flask
python(abi) = 2.7
rpmlib(CompressedFileNames) <= 3.0.4-1
rpmlib(PartialHardlinkSets) <= 4.0.4-1
rpmlib(PayloadFilesHavePrefix) <= 4.0-1
rpmlib(PayloadIsXz) <= 5.2-1
$

It should have at least a few egg dependencies.  I tried to rebuild the package locally and the dependencies were still not found, which would suggest the dependency scanner is still broken somehow.

(In reply to Johnny from comment #3)
> It should have at least a few egg dependencies.  I tried to rebuild the
> package locally and the dependencies were still not found, which would
> suggest the dependency scanner is still broken somehow.

it seems yes,

under Mga4 we have :
Provides

    python-flask = 0.10.1-3.mga4
    pythonegg(2)(flask) = 0.10.1

Requires

    rpmlib(PartialHardlinkSets) <= 4.0.4-1
    rpmlib(PayloadFilesHavePrefix) <= 4.0-1
    rpmlib(CompressedFileNames) <= 3.0.4-1
    python(abi) = 2.7
    pythonegg(2)(itsdangerous) >= 0.21
    pythonegg(2)(jinja2) >= 2.4
    pythonegg(2)(werkzeug) >= 0.7
    rpmlib(PayloadIsXz) <= 5.2-1

under Mag5 we have :
Provides

    python-flask = 0.10.1-6.mga5
    pythonegg(2)(flask) = 0.10.1

Requires

    python(abi) = 2.7
    rpmlib(CompressedFileNames) <= 3.0.4-1
    rpmlib(PartialHardlinkSets) <= 4.0.4-1
    rpmlib(PayloadFilesHavePrefix) <= 4.0-1
    rpmlib(PayloadIsXz) <= 5.2-1


Thierry, any clue ?

Assignee: mageia => thierry.vignaud

The egg generated is correct, so it is our mga5 findrequires process that is broken

$ cat Flask-0.10.1-py2.7.egg-info/requires.txt         
Werkzeug>=0.7
Jinja2>=2.4
itsdangerous>=0.21

Looking at /usr/lib/rpm/fileattrs/pythoneggs.attr I wonder how we even get the provides.

%__pythoneggs_provides  %{_rpmconfigdir}/mageia/pythoneggs.py --provides
%__pythoneggs_requires  %{_rpmconfigdir}/mageia/pythoneggs.py --requires
%__pythoneggs_path      ^((/usr/lib(64)?/python[[:digit:]]\\.[[:digit:]]/.*\\.(py[oc]?|egg.*))|(%{_bindir}/python[[:digit:]]\\.[[:digit:]]))$
%__pythoneggs_magic     [Pp]ython.*(executable|byte-compiled)

So it only wants python executables byte-compiled files.
Maybe this is misleading and it is also called on directories which is why we get provides.

The problem is that pythoneggs.py expects to be called on a file inside that directory, not the directory itself so it does not create properly the Metadata object.

I'll try to patch it.

CC: (none) => pterjan

Committed to git and created a new release.

[pterjan@chopin-cauldron-64 python-flask]$ rpmdiff -iT
/distrib/cauldron/x86_64/media/core/release/python-flask-0.10.1-6.mga5.noarch.rpm
RPMS/noarch/python-flask-0.10.1-6.mga5.noarch.rpm
added       REQUIRES pythonegg(2)(itsdangerous) >= 0.21
added       REQUIRES pythonegg(2)(jinja2) >= 2.4
added       REQUIRES pythonegg(2)(werkzeug) >= 0.7
S.5........ /usr/lib/python2.7/site-packages/flask/helpers.pyc
S.5........ /usr/lib/python2.7/site-packages/flask/helpers.pyo

Thanks!

It may be a good idea to force a rebuild of all standalone python-module packages.  I know at least one other package which has broken dependencies (python-livestreamer, which has a missing dependency on python-singledispatch) but there may be many more.  As I mentioned in my comment #1, I will reinstall the system on my computer as soon as the first beta is released and I will report the missing dependencies I'll stumble upon, but I will obviously be able to test only a fraction of all available packages.  It would be safest to rebuild all module packages in my opinion.

fixed

Status: NEW => RESOLVED
Resolution: (none) => FIXED

I noticed a new package that has a broken python dependency: iep (source package: iep-3.5-1.mga5.src.rpm).  Installing and running âiepâ throws an import exception for the sip module.

$ iep
Started our command server
/usr/bin/python3: Error while finding spec for 'iep.__main__' (<class 'ImportError'>: No module named 'sip'); 'iep' is a package and cannot be directly executed

which is fixed by installing python3-sip.  Should I open a new bug for this or could it be considered this dependency should have been found by the package builder?  The message looks different from usual python import exceptions so my guess is that it is a new bug.

(In reply to Johnny Accot from comment #10)
> I noticed a new package that has a broken python dependency: iep (source
> package: iep-3.5-1.mga5.src.rpm).  Installing and running âiepâ throws an
> import exception for the sip module.
> 
> $ iep
> Started our command server
> /usr/bin/python3: Error while finding spec for 'iep.__main__' (<class
> 'ImportError'>: No module named 'sip'); 'iep' is a package and cannot be
> directly executed
> 
> which is fixed by installing python3-sip.  Should I open a new bug for this
> or could it be considered this dependency should have been found by the
> package builder?  The message looks different from usual python import
> exceptions so my guess is that it is a new bug.

you have to report this against iep package, it is specific to this package

> I noticed a new package that has a broken python dependency: iep (source package: iep-3.5-1.mga5.src.rpm).  Installing and running âiepâ throws an import exception for the sip module.

Thanks for the report Johnny. It should now be fixed in iep-3.5-2.mga5. As Philippe pointed out, don't hesitate to create a specific bug report against the affected package in such cases (well, next time, since hopefully now iep is fixed).

CC: (none) => remi