This is related to #2 - its about installing packages whose setup.py is not at the root level. Not marking as a dupe, though, since #2 is about VCS URLs and this is about tarballs.

I'm curious what algorithm easy_install uses here; does it just walk the tree breadth-first until it finds a setup.py? Probably; guess I could check the code and see...

easy_install just does a glob:

        # Find the setup.py file
        setup_script = os.path.join(setup_base, 'setup.py')

        if not os.path.exists(setup_script):
            setups = glob(os.path.join(setup_base, '*', 'setup.py'))
            if not setups:
                raise DistutilsError(
                    "Couldn't find a setup script in %s" % os.path.abspath(dist_filename)
                )
            if len(setups)>1:
                raise DistutilsError(
                    "Multiple setup scripts in %s" % os.path.abspath(dist_filename)
                )
            setup_script = setups[0]

I might fork and implement, will provide patch then

(I am also 'legalsense')

I'm not fully able to solve this one alone. Started using pip today, maybe developing is a bridge too far. Anyway, partial progress may be interesting for others.

I forked http://bitbucket.org/vanschelven/pip/ and brought in the above mentioned code from easy_install. However, reportlab still fails to install:

Traceback (most recent call last):
  File "<string>", line 14, in <module>
  File "/home/klaas/dev/envs/pip/build/reportlab/ReportLab_2_4/setup.py", line 460, in <module>
    main()
  File "/home/klaas/dev/envs/pip/build/reportlab/ReportLab_2_4/setup.py", line 422, in main
    version=get_version(),
  File "/home/klaas/dev/envs/pip/build/reportlab/ReportLab_2_4/setup.py", line 68, in get_version
    raise ValueError('Cannot determine ReportLab Version')
ValueError: Cannot determine ReportLab Version

It appears that reportlab also tries to read it's own sourcecode to determine the version, and that my patch does not start reportlab's setup.py in the correct cwd (or something similar). Relevant setup.py code:

 38 def get_version():
 39     if daily: return 'daily'
 40     #determine Version
 41     HERE = pkgDir
 42     if os.getcwd()!=HERE:
 43         if __name__=='__main__':
 44             HERE=os.path.dirname(sys.argv[0])
 45         else:
 46             HERE=os.path.dirname(__file__)
 47
 48     #first try source
 49     FN = pjoin(HERE,'src','reportlab','__init__')
 50     try:
 51         for l in open(pjoin(FN+'.py'),'r').readlines():
 52             if l.startswith('Version'):
 53                 exec l.strip()
 54                 return Version
 55     except:
 56         pass
 57
 58     #don't have source, try import
 59     import imp
 60     for desc in ('.pyc', 'rb', 2), ('.pyo', 'rb', 2):
 61         try:
 62             fn = FN+desc[0]
 63             f = open(fn,desc[1])
 64             m = imp.load_module('reportlab',f,fn,desc)
 65             return m.Version
 66         except:
 67             pass
 68     raise ValueError('Cannot determine ReportLab Version')

Any ideas? I'm going with easy_install for now...

Ok, I got my fork to install reportab. Feel free to check it out.

I wouldn't dare to call it a patch yet though, 'cause the tests won't run (at all) with me. I'm getting

IOError: [Errno 2] No such file or directory: './test_basic.txt'

Also, my understanding of the (pip) code is not sufficient to be confident that I'm not breaking anything. I've replaced a number of references to self.source_dir to os.path.basename(self.setup_py), which might have further reaching consequences than I can oversee.

This one really needs a design decision as to whether pip should even support packages who don't put setup.py at the root of the distribution where it ought to be. I'm not convinced that it should: those packages should be fixed.

I agree with Carl on this one. Packages should be fixed to include setup.py in the root of the distro.

For the record, I'm against working around a bug in a 3rd party package.

I agree; I'd at least like to know that the reportlab people have been informed of this problem, and get a response from them, before trying to work around it in pip.

Update or reopen this ticket if this is not for some reason feasible on their end.

Just to make sure I understand the (reportlab's?) problem:

Reportlab's tar.gz as found on Pypi:

/
|--- PKG-INFO
|--- ReportLab_2_4
        |---- setup.py
        |---- src
        |---- ....

Correct would be: PKG-INFO inside that directory "ReportLab_2_4"?

Also: http://www.reportlab.com/ftp/ReportLab_2_4.tar.gz (Reportlab's own download) does not contain any PKG-INFO file.

Any idea how that could have ended up there?

Discussion thread in the reportlab mailing list starts here: http://two.pairlist.net/pipermail/reportlab-users/2010-March/009346.html

I agree that the reportlab tarball on PyPI is broken. A small (but tedious to produce) MANIFEST.in would enable them to produce correct tarballs using the standard mechanism of 'python setup.py sdist', which is the solution I suggested on the list.

(Sorry, didn't notice the login form in the corner of the page. This was me writing comment #11.)

And just to note, if a package has a single directory in it, and a setup.py is in that single directory, everything will work (this is how most packages work). Anything deeper, or if there isn't just one top-level directory, and things start to break.

Verified that as of this morning the reportlab tarball on PyPI is fixed and installs with pip.