Re: GEP 8: Runtime Interface Data Generator For ORbit2

[Owen Taylor <otaylor redhat com>]

Alexander Larsson <alexl@redhat.com> writes:

> On Wed, 25 Sep 2002, Mark McLoughlin wrote:
> 
> > Hey,
> > 	This GEP is a pretty straightforward "Here's what we need and
> > here's the implementation - any objections" type GEP. We'll see how it
> > pans out.
> > 
> > 	http://developer.gnome.org/gep/gep-8.html
> > 
> > 	Discussion is on orbit-list@gnome.org ...
> 
> I don't know very much about ORBit2 (I'm not even subscribed to 
> orbit-list), but this strikes me as suboptimal from a performance 
> perspective. I can just see the profiles of apps starting up and parsing 
> idl files for a long time.
> 
> I would much rather see a solution based on typelibs. And by typelibs I
> don't mean the current shared library thing, I mean architecture 
> independent binary files that can be quickly mmaped and decoded by orbit. 
> Much like the stuff XPCOM and COM uses[1].
> 
> If the IDL compiler generated such typelibs, and there were some automake 
> macros to easily install idl files and build typelibs in 
> $datadir/orbit-typelibs, then they could be read by ORBit at runtime and 
> we would get the same benefits as parsing the IDL files except at a much 
> smaller startup cost.
> 
> [1] http://www.mozilla.org/scriptable/typelib_file.html

The ORBit IDL compiler is actually really fast. I never noticed
it being a bottleneck for CORBA::ORBit compared to the Perl
setup. (Admittedly, I was mostly testing with small IDL files)

And even if there is a precompiled form, it is useful to be
able to compile on the fly and just have the precompiled 
form as an optimization - think of how .py and .pyc works.

Regards,
                                        Owen