Attached are 3 patches to extract the HTTP api as in its own folder/app.

First patch [0001] is moving couch_httpd* in their own application named couch_httpd and make sure tests still pass.

Second [0002] is moving #httpd record in couch_httpd/include/couch_httpd.hrl instead having it in couch_db.hrl.

The third [0003] one create a couch_changes applications (extracting couch changes API from couchdb & couch_httpd applications to remove any circular dependancies and make sure couchdb doesn't use #httpd record. Also by doing this it allows to manage couch_changes API more consistant and usable by other couchdb applications.

All tests pass.

I posted changes on my github repository in the couch_httpd branch :

https://github.com/benoitc/couchdb/compare/master...couch_httpd

bump

I'll try and get a review of this done today. I skimmed and didn't see anything major but I want to go over some of the build stuff and startup in a bit more detail before I can comment more thoroughly.

Commencing brain dump review:

0001

• src/Makefile.am indentation looks wrong for src/couch_httpd

• include files is empty, but this might change in the next patch so just a note to self

• We do a thing for couch.app.src in couchdb so we can set the module list automatically. Should we duplicate that here?

• couch_httpd.app is missing the module to start the app

0002

• Yep, record moved. Looks good.

• Whitespace only hunk at the bottom of 0002

0003

• I'm not entirely certain about this patch. What's the circular dependency? If it's just records in include files I don't think I'd bother making the split as it seems a bit awkward.

• The couch_changes.app reminds me we should be listing dependencies better as we do this. Not super urgent but a thing to think about.

This also brings up how we want to handle things like the httpd code for views. If anyone has good ideas we should open a ticket and discuss a plan of some sort on how to handle this going forward. I'm thinking specifically that the couch_httpd is separate from couch, but couch_changes and couch_mrview have their httpd code in those apps.

Does this ticket make any sense now that we have chttpd, Paul Davis?