Re: Language-independent functions-as-trees representation

[Gabriel Dos Reis <gdr at integrable-solutions dot net>]

Jason Merrill <jason@redhat.com> writes:

| It seems to me that these fundamental design questions need to be
| straightened out now, before we build anything more on top of the current
| IR, or we're just digging ourselves a deeper hole.  As long as we're
| building something new, we should take the time to do it right.  Do others
| disagree? 

All you said make sense to me.  One problem with leaving with history
is that you have to justify why things are the way they are each time
you come across a shortcoming :-)

| I suppose another strategy would be to clean up the expand
| placeholders in the C/C++ IR one by one, rather than try to simplify it to
| something rather different all at once, but that would not address the
| STMT/EXPR issue.

True.  Is it acceptable (from the Release Manager point of view) to
make that change now?  If we can't then it would be a really confusing
situation with the increasing number of branches and the amazing work
already done on them...  If we can, then that may make life a lot
easier... 

-- Gaby