Re: darwin bootstrap failure in mainline

[Dale Johannesen <dalej at apple dot com>]

On Aug 18, 2004, at 10:07 AM, Andrew Pinski wrote:
> On Aug 18, 2004, at 9:54 AM, Fariborz Jahanian wrote:
>
> > Mainline bootstrap fails for apple-ppc-darwin today.
> >
> > ../../gcc-mainline-original/gcc/config/darwin-crt2.c: In function 
> > `__darwin_gcc3_preregister_frame_info':
> > ../../gcc-mainline-original/gcc/config/darwin-crt2.c:146: internal 
> > compiler error: RTL flag check: MEM_READONLY_P used with unexpected 
> > rtx code `reg' in machopic_legitimize_pic_address, at 
> > config/darwin.c:617
>
> This was always wrong so here is the simple fix which does fix at least
> this problem for me.

No, it wasn't wrong, RTX_UNCHANGING_P was defined to be valid on 
registers,
and I'll be surprised if Darwin was the only port doing that.
What exactly it meant was less clear; this code predates me.
It might have been trying to express that the pseudo was set to a value 
once
and never changed thereafter; that seems to fit with the code and the 
doc.
How do we say that now?  I'm concerned about performance impact from
this patch, also; I guess we need to try it.

(btw, there are at least 2 other places in that function that do this, 
Andrew.)

> I am applying it as obvious after double checking this fixes all of the
> problems.
>
> ChangeLog:
> 	* darwin.c (machopic_legitimize_pic_address): Only set MEM_READONLY_P
> 	in the non TARGET_TOC case.