[PATCH] Fix PR debug/19327

[Daniel Berlin]

On Tue, 18 Jan 2005, Eric Botcazou wrote:

>> Just FYI, it didn't introduce the regression.
>> That implies the patch is broken.
>
> Not in my mind.  AFAICT your patch is fine, but it wasn't tested with stabs,
> and caused a testcase that used to pass to fail; so we certainly regressed.

There are only a few platforms with stabs the default, and yes, i 
didn't test it on one of them :)

Note that this actually isn't going to make your optimized debug info much 
better anyway, because you are still going to be missing subblocks of 
blocks not marked used (see the PR that led up to the original patch for 
reasons why).
It just fixes the regression.
If you want to fix the whole problem, you'd also need to change 
dbxout_block to walk subblocks of blocks not marked used, instead of only 
walking subblocks of blocks marked used as it does now.

One of the reasons i didn't touch dbxout and do this when i did the same 
to dwarf2out is because i have no heavy stabs using platforms to test it 
on (darwin doesn't seem to break as easily)
--Dan