Re: [PATCH v2 2/2] Correct invalid assumptions made by (mostly) DWARF-2 tests

["Maciej W. Rozycki" <macro@codesourcery.com>]

On Sun, 16 Nov 2014, Doug Evans wrote:

> >> Sorry for the followup, but FAOD, .4byte.
> >> .word may be a different size on some platforms.
> >
> >  If anything, that would have to be `.4byte'.
> >
> >  I chose `.dc.l' because it is the only fully portable GAS pseudo-op to
> > produce 32-bit data output.  Testing portability was the actual reason
> > to add all the `.dc.*' pseudo-ops to GAS.  As you've already observed
> > `.word' is unportable, and neither is `.4byte' as the latter is only
> > supported for ELF targets.  However in DWARF-2 testing we're probably on
> > an ELF target anyway.
> 
> Yeah.  We've been using .4byte for portability in gdb.dwarf2 as long
> as I can remember.
> 
> I didn't know about .dc.l, but there's nothing in the name that screams
> "portability" to me.  In fact, I can imagine it causing at least a few
> people to have
> to spend time looking it up just to verify it's OK to use (whereas
> they'd be less
> inclined to with ".4byte").  Not good.

 I stuck to the wrong assumption GAS is going to be used with test suite 
runs, as if it was a part of GDB just as it is a part of binutils.  It 
is not, despite sharing the repository.  While maybe less common these 
days another assembler can be used by the compiler run in testing and 
`.dc.l' can only be considered portable among GAS's targets and not 
different assemblers.  So we can't really use the pseudo-op, unlike 
`.4byte' that we already rely on.

> >  So I'll update the tests to use `.4byte'.  Thanks for your review.
> 
> Cool.  I've got one more review coming.

 Applied now, with the update requested and having regression-tested it.  
Thanks again for the review.

  Maciej