Re: XCOFF64 gas patch

[Clinton Popetz <cpopetz at cygnus dot com>]

On Mon, Apr 10, 2000 at 11:02:25AM -0700, Ian Lance Taylor wrote:
 
>    *************** md_apply_fix3 (fixp, valuep, seg)
>    *** 4803,4807 ****
> 		  && operand->bits == 16
> 		  && operand->shift == 0
>    ! 	       && operand->insert == NULL
> 		  && fixp->fx_addsy != NULL
> 		  && ppc_is_toc_sym (fixp->fx_addsy))
>    --- 4844,4848 ----
> 		  && operand->bits == 16
> 		  && operand->shift == 0
>    ! /*	       && operand->insert == NULL*/
> 		  && fixp->fx_addsy != NULL
> 		  && ppc_is_toc_sym (fixp->fx_addsy))
> 
> This is not a good sort of change, because it is not maintainable.
> Either the test for insert == NULL is simply incorrect, and we should
> simply remove it, or there is some more subtle issue here, and we need
> a comment explaining what is going on.

Yikes, this wasn't meant to slip through; I left the comment in to remind
myself to ask about it, but that obviously didn't work.

This is related to the FIXME at the top of that block of code:

         FIXME: We need to handle the DS field at the very least.

because for ppc64 we emit a lot of TOC loads like:

        ld 9,LC..0(2)

which uses:

{ "ld",      DSO(58,0), DS_MASK,        PPC64,          { RT, DS, RA } },

Where DS is:

  { 16, 0, insert_ds, extract_ds, PPC_OPERAND_PARENS | PPC_OPERAND_SIGNED },

So operand->insert is in this case insert_ds, not NULL.  Perhaps someone can
explain why the operand->insert check is there, and I can try to fix it from
there?  

				-Clint