Still bugs in Fortran -fno-emulate-complex

[craig@jcb-sc.com]

>Yeah, that was me.  But Jeff already squashed that one.

Yay Jeff!  :)

>I looked up the archives, it doesn't look to me like there was much
>discussion of MODIFY_EXPR at all.  You said there was a problem with
>Fortran, Mark M said he thought C++ and other front ends might be
>relying on memcpy-like behavior, end of conversation.

Hmm, sorry, IIRC we migrated the discussion to private email, but
I'm not sure we used the right keywords there, either.  (I really
do try to use, and, when designing/coding software, create to use,
keywords pertaining specifically to issues under discussion, because
it *is* so much easier to search for them later.  But it's a difficult
habit to get into.)

>I'd personally vote for adding both MOVE_EXPR and COPY_EXPR, since
>it's not clear which behavior people are relying on.  Then we can
>start migrating the front ends over, slowly.  The whole thing is an
>1.3 project unless it's possible to add tree codes without breaking
>anything - in that case the new codes could go in now.  (Trees aren't
>documented at all; how involved is it to add codes?)

I suspect it'd be pretty easy to add the new codes, #define MODIFY_EXPR
to COPY_EXPR, and let front ends override it as they see fit.  Maybe
I'll look into that -- later, probably not until next week at the
earliest.

        tq vm, (burley)