Consequences for users

• They shouldn’t normally need to see or copy&paste a traceback.
• They will be able to inspect the files before sending them to be sure there’s no sensitive data included.
• As at present, they’ll need a Launchpad account to report bugs in the normal way.

Implementation notes

The use of apport by Bazaar is independent of the configuration in the OS. For example in Ubuntu, apport is normally inactive in release builds, and normally excludes software not installed from a package. We’ll bypass both of them.

Putting in this handler may mean that an OS-wide exception handler never sees the error, but that was true with our existing exception-printer.

The user should have the option to: forget about the crash (and ignore the bug report), see the contents of the report, file a bug, or save the report to file later. At the moment we just show them the filename and let them take it from there.

The process is

1. An exception reaches the top-level handler.
2. We log it in apport-format to a file in ~/.bazaar/crash.
3. We tell the user where that file is, and invite them to file a bug report.

This won’t be active for bugs that cause the whole Python interpreter to crash. This can be handled at the OS level. The nice thing is that if apport is active system-wide, it will catch either exceptions in our in-process apport handler, or errors that crash the intrepreter.

Future ideas

• Capture apport data even for things not believed to be internal errors, because sometimes they are in fact bugs. Then the user can attach the apport report later if they decide to file a bug. There may be quite a lot of them so we might need to limit the number that are stored, or do this when a debug flag is set. At the moment they go into .bzr.log and that’s probably ok to start with.

• Raising an error from the breakin debugger should cause this to fire.

• Developers looking at a crash on their own machine will probably in the first instance just want to see the traceback. Apport files may be more longwinded than our current output and might make the traceback scroll off the screen.

• Automatically trace messages (ie from .bzr.log) in the report. We could just include the whole file, but it may be long, and including the whole thing has a greater risk of including sensitive data.

• Ask the user what they want to do with the report: automatically file it, look at it, see just the traceback, just be told where it is. This could be done through the UIFactory so that it can be done through a graphical dialog.

  However, if we’ve already had an unhandled error in this process there may be problems in Bazaar that prevent us presenting a clean message...

  Possibly these bugs are better reported in the next time bzr runs.