[Distutils] [issue24] Rename easy_install
Phillip J. Eby
pje at telecommunity.com
Tue Jun 17 02:37:18 CEST 2008
At 06:27 PM 6/16/2008 -0500, Dave Peterson wrote:
>Phillip J. Eby wrote:
>>At 12:57 PM 6/16/2008 +1200, Greg Ewing wrote:
>>>Any discussion of that sort seems to get hopelessly
>>>bogged down, so bikeshedding is all that's left for
>>>people to do.
>>
>>What about testing patches? Writing test code? Writing more docs?
>>There's *plenty* of less controversial work to go around. The only
>>reason most of the outstanding patches I'm aware of haven't been
>>applied yet is because they haven't been tested. (Or more
>>precisely, because setuptools hasn't been thoroughly tested with
>>the patches applied.)
>
>
>I can help test some of the patches
Don't test patches - test setuptools with the patches. :) More
precisely, make sure you test things beyond what the patch is
supposed to do, to make sure that other things aren't affected. This
is particularly important for patches to easy_install, which is
ridiculously complicated
due to all the obscure edge cases it has to be able to handle.
> I've seen people posting to the tracker, but I can't commit to
> svn. Beyond running the setuptools test suite
The test suite is pretty useless for most of these kinds of
patches. It essentially only exercises various internals of
pkg_resources and a few other things that are almost never
touched. I'm talking testing as in "actually install some packages
in a few different kinds of install targets, using a few different
options". I don't have a rigorous process for that, as I tend to
pick things on the basis of the code paths to be exercised. But that
might not be an option for casual testers.
If I had it all to do over -- and I didn't simply run screaming from
attempting the task in the first place -- I would write a full
functional test suite, including chrooting tests if necessary. In
the long run, it would have saved enormous amounts of manual test
time, and we could have had more people involved in development a
long time ago.
That, by the way, is why "writing test code" is on the list above.
>and verifying that things seems to work for me and my environment,
>is there anything else that will help get some of the patches into svn?
>
>BTW, most of those things you mention all effectively boil down to
>writing patches in one way or another. :-) How do we make sure that
>after they get some review they get checked in when it seems so few
>people have check-in privileges? Phillip, you already mentioned
>that you're short on time and no one else has responded to a plea
>for finding out who has check-in privileges.
Jim Fulton has previously been "blessed" by me to apply
non-controversial patches to setuptools after giving me a
heads-up. (But note that he's probably busier than I am, and
unlikely to have bandwidth for stuff that doesn't affect zc.buildout
or Zope in some way.)
If you want to expand the available development pool for setuptools,
I would strongly suggest focusing development efforts on creating a
regression test suite emphasizing end-to-end functional testing of
the current functionality. Such tests would ideally be factored for
narrative clarity and compact expressiveness, rather like Jim
Fulton's doctests for easy_install's .exe wrappers, and the doctests
for zc.buildout. (Because if they're too complicated for me to read,
they'll take too long for me to review.)
If people *really* want to solve the development bottleneck, this is
the way to go, because it will reduce my role to deciding whether
something should go in, and seeing if it passes the tests on a
platform or two. That's something I think I could manage to do, even
with my current work load.
More information about the Distutils-SIG
mailing list