slave-only setup.py generates unusable packages

[silk <silk@…>]

Two errors:

1. __init__.py contains: __import__('pkg_resources').get_distribution('Bitten').version so a full bitten still has to be installed.
2. egg-info is missing entry points for recipe_commands

[dfraser]

My bad. Need to think how to handle this...

[osimons]

(Description formatting)

[osimons]

Fix for slave-only install and running builds.

[osimons]

Please try attachment:t383-slave_setup-r663.diff​ - I think it solves both reported issues, and at least my simple 'echo' dummy builds run without issues through a slave-only install.

[osimons]

Updated patch that also adds documentation for the slave-only install.

[osimons]

Actually include all needed files to execute builds, and hide setup() call behind if __name__ == "__main__".

[osimons]

Seems I was a bit tricked by my many test-installs and pre-existing build directory - python setup.py clean --all was needed before running the slave-only install to only get the trimmed fileset included in the install. That showed that none of the build commands were actually included so no builds would execute. All should be there now.

I've also hidden the main setup.setup() call behind a if __name__ == '__main__' conditional to avoid master setup running when recipe_commands are imported. I don't really know if that is supported, but as far as I can tell from my testing with both python setup.py and easy_install it seems to work fine. Anyone know differently?

Finally, I've also removed tests from the slave-only setup as this was not working (exception). I don't see the point in having a test infrastructure for what is actually a single test - a test that is included in main tests anyway.

[osimons]

Slightly more rational - sharing some setup() arguments to avoid duplication.

[osimons]

Committed in [672].

[osimons]

The separate setup scripts don't work so well when we get to distribution - the slave egg will still try to install from the main setup.py.

I've prepared a patch that does this differently, and that also fixes some issues with our beta1 generation of distributions (missing data+++). Starting point at least.

It uses the setuptools.Feature and merges the 2 setup files into one. Basically it enables 'master' by default, but it can be disabled for slave-only build/dist using --witout-master.

There may be some corner-cases I haven't tested, but from my testing this works:

• ./setup.py sdist -> a full sdist
• ./setup.py --without-master sdist -> slave-only, and executable source only
• ./setup.py --without-master test -> runs slave-only tests

I've added testing for existence of MANIFEST.in that was needed so that if for instance a --without-master bdist_egg is run, doing easy_install on this egg later will keep the 'Bitten-Slave' package name and data and not try to re-enable the 'Bitten' information.

Patch follows.

[osimons]

Fresh attempt at solving Bitten vs Bitten-Slave building and distribution.

[osimons]

Changing slave name to BittenSlave + some minor manifest tweaks.

[cmlenz]

Looks good to me!

[dfraser]

A longwinded name - this goes on top of osimon's last patch and prevents slave sdist from having --with-matser

[osimons]

The rework is committed in [758], and will be merged to 0.6 branch shortly.