GC Merge?

[Mark Mitchell <mark at codesourcery dot com>]

Folks --

  I've screwed up.  In particular, as I've continued to work on the
C++ front-end in function-at-a-time mode, I've discovered a number of
problems.  Unfortunately, these problems are already checked in.
Right now, they only affect template functions, which is why the
testsuite has not picked them up.  These problems are serious, and
they need to be solved before the function-at-a-time project can be
completed.

  Most of these problems stem from the fact that obstacks are a
nuisance.  In particular, we were copying things to the permanent
obstack, and some expressions canot be copied.  For example, a
SAVE_EXPR cannot be easily copied in the naive way; if it is
encountered twice it must become the same copy.  I don't think there's
an easy solution, here; I've thought hard about how to do something,
even something temporary, that would solve the problem, and I cannot.
There are other obstack problems as well: it's tricky to make things
end up on the right obstacks in all circumstances.

  Jeff, in a moment of prescience, hinted that this sort of problem
might arise.  I, foolishly, did not quite see the problems.  Now I do.

  In an effort to redeem myself, I intend to get RTH's GC work working
on the mainline, and get it merged in.  Of course, I need approval to
do that.  

  Jeff, Richard, do you recall any particular design problems with the
current GC implementation?  In other words, short of run-of-the-mill
bugs, what were the reasons this patch was never incoprorated?  

  Are there any objections to my plan?

--
Mark Mitchell                   mark@codesourcery.com
CodeSourcery, LLC               http://www.codesourcery.com