On Wed, 18 Nov 2009 22:02:31 -0800, Andreas Jung <[email protected]> wrote:
Am 19.11.09 00:42, schrieb Robert Kern:
Personally, I am entirely uninterested in moving up grades. I am
interested in having good, discoverable documentation,
Amen. Any PyPI package release w/o proper metadata and without
reasonable description/documentation
is a broken release and should be banned from PyPI. Package quality
starts with your metadata and
the willingness of a package maintainer fulfilling certain minimum
standards.
I agree about metadata (not sure about documentation). Based on what I see
from building packages[1] for PyPM, most packages fail due to one of the
following reasons:
1) Missing PKG-INFO file (the author did not use the `sdist` command).
Twisted, IMDBPy are some examples.
2) Trying to read a non-existent file from setup.py (eg: author forgot to
include README.txt in the tarball -- buggy MANIFEST.in?)
3) no setup.py
4) reading stdin in setup.py (so the "setup.py build" runs indefinitely)
5) no downloads URL (no tarballs either)
7) Import itself in setup.py (foo-0.1.tar.gz/setup.py has "import foo" --
and that in turns imports uninstalled deps)
6) Missing "build dependencies" (many packages try to import
numpy.distutils/twisted so on)
Other failures usually include missing library dependencies (libxml, for
instance) or some Python syntax error.
-srid
PS: Now that we have the build infrastructure that periodically (i.e.,
every day) builds packages from PyPI, I might experiment with measuring
the core "installability" rating for all packages sometimes during the
weekend.
***
[1] reports at http://pypm.activestate.com/
_______________________________________________
Catalog-SIG mailing list
[email protected]
http://mail.python.org/mailman/listinfo/catalog-sig
