Re: introspection for 2.32

[Colin Walters <walters verbum org>]

On Fri, Sep 10, 2010 at 9:29 AM, Vincent Untz <vuntz gnome org> wrote:
>
>> Is it just a matter of rebuilding the packages that have introspection
>> support? Or will they break in weird ways because of that change?

The complication comes from the fact that before, we only had a
--strip-prefix option for the scanner.  The semantics of that were
messed up; in 0.9.5 we added two new different options,
--identifier-prefix and --symbol-prefix which are a lot clearer and
supportable.

We also added --include-uninstalled for the complicated case of
Clutter, which actually builds 4 different GIR files from the same
.so, with dependencies between them.  Previously clutter+g-i
effectively BuildRequired itself because the scanner would be looking
in the installed tree...we didn't really notice because the normal
jhbuild workflow has previously-installed versions, but of course OS
vendor build processes don't (but this doesn't get tested until very
late in the release process).

I've then been updating modules to use this by having them require g-i
0.9.5.  Thus currently it's 0.9.5 or nothing.  Now, with potentially
some implementation work, I could try to readd support for the build
API added since then to 0.9.3.  But it's not trivial, and I don't know
if I could deliver that for 2.32 realistically.

My recommendation is basically that 2.32 distributors take a choice of
0.9.5 or nothing; for the latter, in practice it's just a few
gnome-games.  I don't think we're planning to do a GNOME 3 preview
release for 2.32 - but if we were, we could require g-i.

At least in Fedora I plan to try pushing through 0.9.5, because
basically 0.9.3 was so rickety and full of large-scale issues (not
trivial bugs) that it prompted me to take a month to rewrite it as
mentioned in the status report.