http://git.collabora.co.uk/?p=user/danni/telepathy-glib.git;a=shortlog;h=refs/heads/g-i-fixes

g-ir-scanner can now give warnings. The second patch fixes up some of these warnings.

> -       _gen/telepathy-interfaces.h

Why?

> +       $(srcdir)/connection-handles.c $(srcdir)/connection-contact-info.c \

Have you checked that everything in connection-contact-info.c has appropriate g-i?

Other than that, looks good.

> +GOBJECT_INTROSPECTION_CHECK([0.9.5])

I hope this is excessive: no distribution tracked by whohas(1), including Ubuntu and Fedora, has newer than 0.9.3 at the moment (and Debian 6 will have 0.6.14).

Given that g-i appears to be a hard dependency, I'm unhappy about bumping the version to one considerably newer than what's in distributions.

It would be ideal if we could keep compatibility with both 0.6.14 and 0.9.5 (if the only difference is that you pass different command-line options, that doesn't look like an insurmountable problem).

If we really can't be compatible with 0.6.14 at all, we should make g-i optional.

(In reply to comment #2)
> > -       _gen/telepathy-interfaces.h
> 
> Why?

It's being included by interfaces.h and seems to break the new scanner to have the file scanned twice. It's worth noting that the new scanner seems to ignore string #defines at the moment.

> > +       $(srcdir)/connection-handles.c $(srcdir)/connection-contact-info.c \
> 
> Have you checked that everything in connection-contact-info.c has appropriate
> g-i?

Yeah, it looks annotated, but it was left out of the scan.

(In reply to comment #3)
> > +GOBJECT_INTROSPECTION_CHECK([0.9.5])
> 
> I hope this is excessive: no distribution tracked by whohas(1), including
> Ubuntu and Fedora, has newer than 0.9.3 at the moment (and Debian 6 will have
> 0.6.14).

It probably is excessive. I'm having trouble working out what the precise version required is. They don't believe in release tagging :-(

> Given that g-i appears to be a hard dependency, I'm unhappy about bumping the
> version to one considerably newer than what's in distributions.

Should we perhaps be using GOBJECT_INTROSPECTION_REQUIRE() then?

> It would be ideal if we could keep compatibility with both 0.6.14 and 0.9.5 (if
> the only difference is that you pass different command-line options, that
> doesn't look like an insurmountable problem).

This seemed easy enough. I have added an (untested on 0.6.14) patch that should allow us to use either version, that we can drop at some point in the future.

*** Bug 30133 has been marked as a duplicate of this bug. ***

Stable branch blocker, since it causes FTBFS with newer g-i.

Note that --strip-prefix has been forward-ported for compatibility reasons:

http://git.gnome.org/browse/gobject-introspection/commit/?id=920512de3fb9bfefffadb454308d66eca9c9ebd4

Stealing this to make sure it's in 0.12; see my g-i-fixes branch.

(In reply to comment #5)
> > Given that g-i appears to be a hard dependency

On closer inspection, it isn't, leading to Bug #30134 (which my branch also fixes). 

> I have added an (untested on 0.6.14) patch

Works for me (I'm running 0.6.14 on Debian).

I'd prefer it if you combined the command-line arguments for the two cases as much as possible, though; my branch does so.

Fixed in git for 0.11.15