Re: [Patch, _eh, dawin, Version2] Allow targets to suppress epilogues in _eh frames.

[IainS <developer at sandoe-acoustics dot co dot uk>]

On 16 Aug 2010, at 18:44, Richard Henderson wrote:

> On 08/16/2010 09:38 AM, IainS wrote:
> > On 16 Aug 2010, at 17:18, Richard Henderson wrote:
> >
> > <snipped > .... other points taken on board....
> >
> > > Also, I'm noticing that dwarf2out.c uses
> > > flag_asynchronous_unwind_tables in places where it really means
> > > cfun->can_throw_non_call_exceptions. In fact, almost all
> > > occurrences of f_a_u_t in dwarf2out.c are in error. The only
> > > exception that I can see off-hand is in fde_needed_for_eh_p, which
> > > would need an extra check vs cfun->can_throw_non_call_exceptions.
> >
> > How should we proceed on this?
> Patch dwarf2out.c?  How else?

heh,
I guess I meant was solving the problem you noticed above a pre- 
condition of applying the patch, or should they be considered separate  
problems?

> > And also there is the remaining question about the GNU-specific code
> > (which I fear will get re-enabled when the stuff above is fixed).
>
> What remaining question?  What GNU-specific code?

this....

static void
dwarf2out_args_size (const char *label, HOST_WIDE_INT size)
{
  dw_cfi_ref cfi;

  if (size == old_args_size)
    return;

  old_args_size = size;

  cfi = new_cfi ();
  cfi->dw_cfi_opc = DW_CFA_GNU_args_size;  <<<============
  cfi->dw_cfi_oprnd1.dw_cfi_offset = size;
  add_fde_cfi (label, cfi);
}

is a dwarf-2 target supposed to respond to this (i.e. we have a darwin  
bug if it doesn't work) or
... should there be a fall-back for dwarf-2 strict?

=====

I'm attaching a consolidated patch which takes on board the comments  
you've made.
It regtests on i686-darwin9 and java functions on darwin10 with it in  
place, the same patch also restores Java functionality on 4.5.2.

(I've left two debug prints in place for now - accepting that they  
need to be removed before it's applied).

I'll cross-reference this in PR41991 in case someone else wants to  
pick it up but, unfortunately, I don't have time to advance it at the  
moment.

cheers,
Iain

Attachment: 163302-unwind-epilogues-v2a-diff.txt
Description: Text document