-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1 Jim Fulton wrote:
On Nov 15, 2007, at 9:35 AM, Tres Seaver wrote: ...
Chris is not advocating switching away from zc.buildout for the *development* of the packages; he is arguing that we can't force people who *use* the packages to stop using the facilities supplied setuptools.
Agreed,
If we want people outside the rather small group of core Zope developers to *use* our software, we are going to need to accomodate their expectations about how Python pacakges work. At a miniumum, any package which we release to the Cheeseshop should be testable and installable using the "standard" tool used by nearly everyone: setuptools.
That means we need to work harder at a coule of things:
- Making sure the install-time dependencies in setup.py are accurate, minimal, and themselves installable.
Yup
- Making sure that the tests can psss for somebody who is not a core Zope developer, (and therefore who doesn't know about zc.buildout, or want to). E.g.:
$ /path/to/python setup.py develop $ /path/to/python setup.py test
Grrrrr. I find "setup.py test" to be really annoying as it requires meta data that is available only to that command.
Not after running 'egg_info', or any other setuptools-aware command; the tests_require dependencies then become introspectable. E.g.: $ svn co $ZSVN/zope.tal/trunk zope.tal-trunk $ cd zope.tal-trunk $ /path/to/setuptools-aware/python setup.py egg_info $ cat src/zope.tal-egg-info/requires.txt setuptools zope.i18nmessageid zope.interface [test] zope.testing
If there are dependencies which are needed to run the tests but not merely to install the package, there should be a way to spell that.
Agreed. I'll note that there is a school of thought that I proposed but am somewhat ambivalent about that anything required by tests should also be required by install. This is in accordance with "fly what you test, test what you fly." To the degree that we do this, then there is no need for tests_require.
There are a bunch of packages out there which contain ZCML, and try to test the loadability of it. Those tests end up creating a huge furball of dependencies which are artificial: the package itself doesn't need to process ZCML, and any application which wants to conume the ZCML provided by the package is going to have those dependencies satisfied anyway.
In setuptools, that is what 'tests_require' is for; buildout doesn't (yet) have such a spelling (I think).
Ideally, buildout (really the testrunner recipe) should be able to honor tests_require, but setuptools doesn't make this information available to any other packages. :( (big frowny) Also, I believe that more is required to get "setup.py test" to work. I'm unsure what the details are.
See above. If buildout can trigger creation of the egg-info directory, all should be well. In fact, that directory will already be present in any source distribution built with setuptools.
When we do this, we tend to use a test extra. I suggest that when we use a test extra, we also put the value in tests_require. I will find this so distasteful that it will help me stick to the strategy of just putting these dependencies in install_requires. :)
In order to succeed here, we are going to have to quit releasing packages without more careful thought, testing, etc.
+1
This should not be interpreted as criticism of the heroic and well- meaning efforts to date. Despite that inevitable mistakes that were made, I'm very appreciative of the efforts a few individual people made to make the transition to eggs possible. I also apologize for not providing more oversight.
Agreed completely (on the heroic part). Tres. - -- =================================================================== Tres Seaver +1 540-429-0999 tseaver@palladion.com Palladion Software "Excellence by Design" http://palladion.com -----BEGIN PGP SIGNATURE----- Version: GnuPG v1.4.6 (GNU/Linux) Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org iD8DBQFHPGtF+gerLs4ltQ4RAiflAKCjtH9ZutX/rL3v4w3qC6antxWUMQCfWgmU AFNKLCpInKK9xNB3U2Aliws= =ooWa -----END PGP SIGNATURE-----