Config pkg, take 2 - #73

@mdboom - Thanks for the detailed feedback! Everything that's not commented on above should be fixed in the version I pushed up just now.

Regarding the auto-writing of the first file: I'm very much against the by-hand approach, because I'm certain there will be lots of problems with people forgetting to do so. I tried to write this system here so that the output config file is still pretty nice looking with the necessary comments and so on.

So yeah, I think the best approach is to do something like setup.py test does - my thought was that this could be a new setup.py command, allowing people to wipe their current config and startover if they provide the appropriate option, but otherwise it only creates a new one if it isn't there already. Then we could set it up so that the setup.py install and setup.py develop commands run that command (in the non-overwriting form) when they finish everything else.

@astrofrog,@iguananaut - you two both had a variety of comments on the last pull request, so I'll wait for you two to either give feedback or say it looks good before merging anything.

@hamogu, @phn - you may also be interested, as you both had a few comments in the old pull request.

This are notes mostly for myself, as I will make issues for each of the not-yet-completed items after this pull request is done:

• Provide some sort of progress feedback when downloading remote data
• implement data.astropy.org server
• ?move the function that locates it's caller's module to utils?
• ?unified file-locking scheme?

I haven't read everything in detail, but it seems fine to me. In any case, I'm sure we will need to tweak things once we start actually making use of the config utilities.

@eteq: As for writing the defaults -- I think I wasn't clear. setup.py can not be responsible for writing the file to the user's home directory. The person running setup.py may be a sysadmin installing on a shared server, which is then run by users in multiple locations. These users probably don't have access to setup.py or at best won't know where to look for it.

What I meant to say is that setup.py build (why make it an additional step?) would create the template file, and install that along with astropy. Some mechanism on importing astropy once installed would check the user's home directory, and if finding no config file, copy the template there. We can also provide an easy-to-find function to reset all configuration files, too.

A simpler solution would be to not generate and install a template file, but simply generate it when needed. This has the downside that all of astropy needs to be imported upon first usage -- maybe not a big deal at present, though.

Looks good to me. Once this is merged, there are some places I'd like to start using it :)

@mdboom - ah, I get what you mean now. Given the range of options there, I'll make sure to put it up as an issue and there can be a later pull request to deal with how to actually organize the generation.

The only problem I see with this is that autogenerating a config file means every module has to be imported once to make sure all config options are found. I don't think there's a way around that though. Honestly, I'm not against including a template config file (or files) in the source code--if a new option is added it would just have to be added to the template. There could even be code to generate it.

On the other hand, as long as the generation only has to happen once that's fine too. There should be a function (if there isn't already) to update a user's config file with new options while preserving existing settings.

All the remaining changes seem to be targets for later pull requests, so I'll merge this now. Thanks for all the helpful feedback!