On Wed, Feb 14, 2001 at 08:46:45AM +1030, Berndt Josef Wulf wrote:
> Thor Lancelot Simon wrote
> > On Tue, Feb 13, 2001 at 12:48:25AM +1030, Berndt Josef Wulf wrote:
> > > Alistair Crooks wrote
> > > > > 
> > > > > --- Makefile.orig	Tue Feb 13 00:07:03 2001
> > > > > +++ Makefile	Tue Feb 13 00:01:19 2001
> > > > > @@ -10,7 +10,7 @@
> > > > >  MAKEFILE=		GNUmakefile
> > > > >  
> > > > >  # Needs ELF clue, mips has no TAS implementation
> > > > > -NOT_FOR_PLATFORM=	*-*-alpha *-*-mips
> > > > > +#NOT_FOR_PLATFORM=	*-*-alpha *-*-mips
> > > >
> > > > I'd prefer to keep the "NOT_FOR_PLATFORM= *-*-mips" in there,
> > > > especially if the patch is to be "applied for NetBSD-alpha
> > > > architectures only..."
> > > 
> > > Actually, it would be interesting to see if it compiles on mips...
> > 
> > No, it wouldn't!  Unless you consider it similarly interesting to
> > determine that it *works correctly* on mips.  Do you even understand
> > what the missing code was supposed to *do*?
> 
> I thought I did... and no, I don't know the MIPS archicture. Good to
> see someone who does.

Hey -- I didn't mean to be so rude, apologies.

But there have been a number of "patches" that made various bits of
pkgsrc that wanted test-and-set functionality "work" on various platforms
by patching out the calls to the missing test-and-set functions/macros!

Obviously, that is not a good idea.

Test-and-set on MIPS is hard, since the earlier processors don't have an
atomic TAS or compare-and-swap instruction.  I'd want to be _really_
sure that the implementation was actually bulletproof before turning
on the build of a package #'d out for that reason on MIPS -- just
compiling doesn't seem good enough to me.