[Patch, Fortran] PR42484: ICE with -fopenmp

[Jakub Jelinek]

On Mon, Dec 28, 2009 at 10:38:47PM +0100, Janus Weil wrote:
> 2009/12/28 Jakub Jelinek <jakub@redhat.com>:
> > On Mon, Dec 28, 2009 at 09:42:54PM +0100, Janus Weil wrote:
> >> here is my patch for this PR. It took me a while to find the root of
> >> the ICE (for history see the PR log).
> >
> > This is wrong
> 
> What is wrong? Can you be a bit more specific?

The patch was a workaround, rather than a real fix.  We shouldn't be adding
workarounds, but should fix the real underlying problems.

AFAIK we don't emit DW_AT_decl_column/DW_AT_call_column in the debug info
yet, so the only reason to add precise column information to location in FE
is diagnostics, but in the Fortran FE where the majority of diagnostics is
emitted from the gfortran's private location representation using caret
diagnostics etc. this doesn't make any sense.

Being able to emit a GIMPLE_COND instead of GIMPLE_COND and a bunch of
GIMPLE_GOTOs is desirable, the C/C++ don't do that just to make sure the
debuginfo for say
if (something)
  goto lab1;
else
  goto lab2;
allows putting a breakpoint at either of the goto lines, at least with -O0
-g.

	Jakub