Sorry to bring up an old issue, but I just got bit by this when trying
to install a python package.

The `config` command was used to ensure that a certain function was
included in the math.h on my cpu before compiling the resulting c binary
and python wrapper for it.

I had to chase down the problem and basically re-figure out what Michiel
did in his patch on my system[1] and override the `try_run` method in
the setup.py file.

I didn't bother checking the issue tracker assuming distutils is used so
widely (and is old enough) that the bug couldn't be in distutiles
itself, but must be something I'm doing wrong in my end ... oh well.

Anyway, could we somehow try to fix this bug to mitigate some lost time
and frustration?

Michiel's patch seems reasonable enough to me since it looks to be the
least surgery-intensive way (I'm not sure where Phillip would like to
put the os.path.abspath() call) to fix the bug for the majority of the
time when someone would be bitten by it.

Regardless of whether or not the code is well documented, it's still
there and usable in distutils (w/o any warnings against its use), so it
would make sense to me that we at least have it work as its intended to.

[1] My path was slightly different than the submitted patch. Where
Michiel has:

command = os.path.join(os.curdir, exe)

I have:

command = os.path.join(os.getcwd(), exe)

all in all, you get the same effect