Class initialization, take two

[Tom Tromey]

>>>>> "Jeff" == Jeff Sturm <jsturm@one-point.com> writes:

Jeff> My last classloader patch inadvertently broke Kawa builds.  This
Jeff> corrects it and also improves support for -fno-assume-compiled.

I'm sorry I've neglected this so long.  I've been single-mindedly
focused on something else.  I'm hoping to get through the patch
backlog over the next few days.

This patch looks weird, since it moves things around in a way I wasn't
expecting.  However, I also ran into some new class init bugs, and
this patch definitely fixes them.

So, I think this should go in on the trunk.

If you're playing with -fno-assume-compiled, there are some other bugs
in that area that I've fixed in the runtime.  I'll get those fixes in
shortly.

Jeff> This patch is one way to handle it.  An alternative is to add an
Jeff> explicit link step to class initialization, as the bytecode
Jeff> interpreter does.

Do you think this would yield a cleaner result?  I've been in denial
about parts of class initialization for years now.  Lately, though, I
find I can't ignore it any longer.  We have a lot of different
functions, all laid out rather confusingly, sometimes with overlapping
functionality.  It would be nice to clean it all up.  I'll submit a PR
to that effect.

Tom