Re: [m68k 12/13] properly align within code

[Jeffrey Law <law at redhat dot com>]

On Tue, 2007-02-13 at 15:03 +0100, Roman Zippel wrote:
> Hi,
> 
> On Mon, 12 Feb 2007, Jeffrey Law wrote:
> 
> > On Wed, 2007-02-07 at 22:09 +0100, Roman Zippel wrote:
> > 
> > > > Assuming that's the case my only concern would be whether or not
> > > > the .balign pseudo-op is accepted by the various m68k assemblers.
> > You've made that assertion, but how do you know that?  All the the
> > world is not gnu.
> 
> As I've shown proviously at least the m68k world has shrunk considerably.
I'm not doubting that.  The question is has the world shrunk enough that
you can simply remove support for an assembler variant that we have
supported for 20 years.

> 
> > > Currently gas is the only assembler left and it's only generated if one 
> > > uses one of -falign options.
> > But again, this hinges on your assumption that the only m68k assembler
> > in use is gas.  Just asserting that gas is the only m68k assembler in 
> > use isn't enough, you've got to convince us that's the case since this
> > kind of patch would break support for other assemblers.
> 
> Sorry, but how am I supposed to "prove" that???
You can't prove it, but just asserting it isn't sufficient.  You
have to find a middle ground.

One thing you could do that would remove my objection would be to
show me that there's no way the MOTOROLA_SYNTAX code could possibly
work in the previous GCC release.  Or that there's no way to actually
make GCC produce MOTOROLA_SYNTAX in the previous GCC release.

If that's not possible, then that gives you a direction you can 
legitimately take.  Disable MOTOROLA_SYNTAX, but keep the code
working for a release.  When there aren't any complaints after
a suitable period of time after the next release, then zap all
MOTOROLA_SYNTAX support.

> If nobody speaks up in support of other assembler, I don't see any value 
> in keeping useless code. At the remote chance there is still someone out 
> there trying to use gcc with an alternative assembler, we likely won't 
> hear about it until it breaks somehow and if that should happen, I'd 
> consider it a good thing, so we actually _know_ what kind of support is 
> needed. OTOH I'd consider it a real bad thing, if this fear of the unknown 
> kept us of doing some very real and actually useful fixes.
And the way to get to this state is to disable it, but not otherwise
break it and wait for a period of time to give the end users a chance
to complain.

Jeff