Re: Binary Compatibility ABI work

[Andrew Haley <aph at redhat dot com>]

Jeff Sturm writes:
 > On Sat, 28 Feb 2004, Andrew Haley wrote:
 > > However, some folk have got to hear about this work and they might
 > > think I'm being unnecessarily secretive.
 > 
 > That could be my fault, having asked about it on irc.
 > 
 > I didn't mean to imply anyone was secretive, just that it's difficult for
 > anyone to help given the current state of the project.  Even though it's
 > hard for me to commit time right now, this work is of vital importance to
 > gcj, and I want to help, even if I can only do regular builds and
 > testing.

That is a reasonable point of view, sure.

 > > To counter this appearance,
 > > I will check in the patch in its current state on a branch if people
 > > wish.
 > 
 > If a branch is chosen for this work in progress, I can also offer to help
 > with merges and any CVS operations needed (that's part of my day job!).
 > 
 > Are any runtime patches needed yet, or just gcc/java?

Quite a few, actually.  In the process of doing this stuff, I have
discovered a number of race conditions in libgcj where Thing A needs
to be initialized before Thing B which in turn depends on thing A.
Argh.

We didn't notice this before becasue the pre-BC compiler statically
resolves everything.  I haven't solved these, but just kludged around
them in order to make stuff work.

The real secret to achieving binary compatibility is to resolve
bindings as late as possible without incurring perfomance penalities.
This is hard.

 > 
 > > My natural reaction as a professional prgrammer is to say "Dear God,
 > > don't let anyone see this."
 > 
 > *grin*  I think every professional has written code like that.
 > 
 > Among the things I really respect about the free software process is the
 > qualify of code contributed.  (Much more so than certain proprietary
 > projects I've witnessed over the years!)

I hope so!  Well, that's the plan, anyway...

Andrew.