This is the mail archive of the
gcc@gcc.gnu.org
mailing list for the GCC project.
Re: Two problems with gimplify_init_constructor
- From: kenner at vlsi1 dot ultra dot nyu dot edu (Richard Kenner)
- To: rth at redhat dot com
- Cc: gcc at gcc dot gnu dot org
- Date: Thu, 12 Aug 04 16:42:01 EDT
- Subject: Re: Two problems with gimplify_init_constructor
> Actually, I meant suppress the clearing operation in the readonly case.
Then no, we can't do that.
Why? Isn't it just an optimization? It certainly is in this case?
(Actually, it won't get done at all if elements are counted right. Changing
that will "fix" this bug, but will really make it latent.)
We should be suppressing whatever we have to suppress to get
that damned RTX_UNCHANGING_P bit off.
We can't, though. In this case, the LHS is an INDIRECT_REF whose type is
a record with a field that's readonly. When you have a readony field,
it'll be read with /u, so a store to the entire object needs /u and that's
the problem. So if we *could* somehow get it off, we'd have another problem.
I swear I'm gonna rip it out one of these days...
That time is indeed coming closer and closer ...