[PATCH] another discarded DWARF issue

[Bob Wilson]

On Oct 15, 2007, at 6:22 PM, Alan Modra wrote:

> On Tue, Oct 16, 2007 at 10:02:17AM +0930, Alan Modra wrote:
>> On Mon, Oct 15, 2007 at 03:42:37PM -0700, Bob Wilson wrote:
>>> -      if (kept != NULL && sec->size != kept->size)
>>> +      /* The kept section may have been relaxed; compare its  
>>> original size.  */
>>> +      if (kept != NULL
>>> +	  && (elf_section_data (sec)->this_hdr.sh_size
>>> +	      != elf_section_data (kept)->this_hdr.sh_size))
>>
>> What if "sec" is non-ELF?  (I'm not dead certain this can happen, but
>> I think it can..)

I'll take your word for it.  I wasn't sure how to do this.  See below.

>
> I think you should probably use
>
>      if (kept != NULL
> 	 && ((sec->rawsize != 0 ? sec->rawsize : sec->size)
> 	     == (kept->rawsize != 0 ? kept->rawsize : kept->size)))
>
> OK with that change.

I did that at first but then read the comment describing the rawsize  
field: "it is currently only set for sections where the linker  
relaxation scheme doesn't cache altered section and reloc contents  
(stabs, eh_frame, SEC_MERGE, some coff relaxing targets), and thus  
the original size needs to be kept to read the section multiple times."

The Xtensa port _does_ cache the altered section contents and  
currently does not set rawsize.  I started changing it to set rawsize  
but this comment made me think that perhaps that was the wrong  
approach.  I guess not, since there doesn't seem to be another good  
way to get the unrelaxed size of a section.  Unless you have other  
suggestions, I'll apply the patch with your change, and then fix up  
the Xtensa port to set rawsize.

Thanks,
Bob