[Libvir] PATCH: Cleanup generator & force build fail on missing APIs

[Daniel P. Berrange]

On Mon, Jan 21, 2008 at 04:14:15AM -0500, Daniel Veillard wrote:
> On Sun, Jan 20, 2008 at 05:29:06PM +0000, Daniel P. Berrange wrote:
> > The python generator will happily ignore functions it can't handle and
> > pretend everything completed without error. This leads to the situation
> > where we add new APis to C library and no one ever notices that they
> > are missing from the python until months later. This requires that my
> > previous patch be applied first to implement the missing APIs we already
> > have :-)
> > 
> > This patch causes the generator to return a non-zero exit status if there
> > are any APIs marked as FAILED. It will also explicitly print out their
> > names so its clear what is missing. In doing this I added a bunch more
> > functions to the skip list - ones that we already manually wrote.
> > 
> > It also removes the manually written virCloseConnect/virDomainFree/
> > virNetworkFree C code, since the generated code is just fine.
> > 
> > Finally, it makes all manually written C functions static for consistency
> 
>   Okay, the default behaviour prints the number of functions which failed
> (and the number skipped) but now that we have full coverage, yes this is
> a good thing to do. Note however that you're likely to still see the problem
> of late discovery of missing bindings because:
>   - people submitting patches are likely to just run 'make'
>   - people applying it will do the same.
>   - only on 'make rebuild' in docs or when preparing the release
>     will we hit the docs/libvirt-api.xml , leading to the subsequent
>     error on a missing part.
>   - and I'm afraid I will be the one hitting them ... at time of release
>     i.e. at the worse moment with a make exiting on an error.

Never fear - this is exactly the sort of problem the nightly autobuild
is intended to catch - it'll fail the night after commit and send an 
email alert so we can fix it the very next day - hopefully long before
release.

>   So okay to fail, but only if 'make' leads to a failure after modifying the
> API headers, which probably means changing docs/Makefile.am to systematically
> rebuild when libvirt.h (or .h.in since libvirt.h is autogenerated ?). Or some
> other mechanism which somehow force people changing the API to rebuild the
> XML and the bindings.

IMHO the value of it is that it always runs & fails - causing an immediate
failure will ensure that it gets addressed immediately and not ignored /
undiscovered for months.

Dan.
-- 
|=- Red Hat, Engineering, Emerging Technologies, Boston.  +1 978 392 2496 -=|
|=-           Perl modules: http://search.cpan.org/~danberr/              -=|
|=-               Projects: http://freshmeat.net/~danielpb/               -=|
|=-  GnuPG: 7D3B9505   F3C9 553F A1DA 4AC2 5648 23C1 B3DF F742 7D3B 9505  -=|