CodeHosting server should generate OOPSes

this means getting the sftp server to record errors as oopses?

That's the idea. If anything goes wrong during the life of a session, it should record an oops instead of logging the error in a log file nobody looks at.

An examples of known errors we had were:
 * the xmlrpc client had not been updated, and there was an error at the of the session.
 * the vfs directory observer mechanism was buggy and prevented users to create new directories under .bzr

Both are fixed now, but there might be other problems we are not aware of. Generally, oopses should record two sorts of errors:

 * uncaught exceptions
 * user-visible errors (sftp protocol, or authentication)

We can start with just "uncaught exceptions" and "error reported to the client", and then silence out oopses for errors that are produced as part of the normal operation of bzr dumb client (such as attempting to create already existing hash directories in ./bzr/repo/knits).

If would be nice if the "uncaught exception" reporting could be implemented as a reactor feature so it would be truly global.

Assigning to jml. Do you need more info?

I don't think we should OOPS for all user-visible errors. OOPSing for uncaught exceptions seems fine. Is there a policy on OOPS usage?

I'm pretty sure that Twisted has a handle for uncaught exceptions. At worst, we can hook into the logging system and record an OOPS for any logged errors. I think doing this + tests should constitute fixing this ticket.

I'm not getting this done before I go on leave.

Well, there's something at bzr+ssh://devpad.canonical.com/code/mwh/launchpad/codehosting-oopses-bug-90484. Comments please.

Not going to make 1.1.12 because I haven't figured out how to test it yet.

Fixed in RF 5639.