[qmtest] running qmtest on bash in cygwin...

[Stefan Seefeld]

Mark Mitchell wrote:

 >>just from looking at the GNUMakefile, I see that the version of python
 >>being used decides also whether to install the qmtest (cygwin shell)
 >>script or the qmtest.bat batch file. Does this really make sense ?
 >
 >
 > I'm not sure.  We don't want to key off of config.guess output, because
 > that will always indicate Cygwin, even when using the native Win32
 > Python.  Ideally, we'd like to get rid of the shell script/batch file
 > that is used to bootstrap QMTest, but that's another project...

...speaking of which: I had a closer look at the current build procedure
qmtest is using. I also had a little discussion with Nathaniel concerning
the specific requirements this build system should meet.

What about a build procedure purely based on python's distutils package ?
The setup.py scripts can contain different sets of files to install:

* python modules
* data files
* scripts

And with 'python setup.py --prefix=...' or 'python setup.py --home=...' you
have pretty much complete control over where everything is installed. There
are even more fine-grained options to specifically control scripts, data files,
and modules, if the above isn't sufficient.

The default installation would put the modules under python's site-packages,
and the rest into /usr/local/[share, bin], which is what I would expect
from a typical debian / rpm package.

And for those who want everything in a single isolated place, an installation
using the '--home' parameter would probably do the trick. In any case, the
main executable could be a python script, and some little experimentation
suggests that the qm/test/qmtest.py file is almost ready to serve, i.e.
the qmtest.[bat, sh] files aren't really needed then.
There are only two parameters: the PYTHONPATH, if the modules are not
in the default location, and a variable (which could be substituted into
an installed module) that indicates the location of the data files.
That's about it.

Am I missing anything ?
I would volunteer to provide the needed patches, if you agree that this
is the right thing to do...

Kind regards,
		Stefan

PS: to be able to execute qmtest from within the working directory the cvs
     repo file layout should probably be adjusted a little, but that is, as
     njs told me, something you'd consider in a cleanup anyways ;-)