[llvm-commits] [llvm] r107625 - in /llvm/trunk/lib/Target/X86: X86ISelLowering.cpp X86Subtarget.cpp X86Subtarget.h

[Duncan Sands]

On 05/07/10 21:54, Anton Korobeynikov wrote:
>> If not we should fix it in triple. (and for now just "don't do that" if it is a problem)
> Well, Daniel one day was strongly opposed to this. He thought that
> such "short triplets" should not exist...

I never really understood Daniel's argument.  Clearly we should not *produce*
bogus triples, but I think we should *accept* odd "triples" that people pass to
us.  People (including major distributions) regularly configure gcc etc with
non-normalized triples, so correctly handling them seems important.  It's not
surprising that people get this wrong: the web is filled with contradictory
descriptions of how triples are "supposed" to be [*], not to mention examples
of wild triples with the components in any order etc.

Ciao,

Duncan.

[*] The definitive description seems to be: a correct triple is what is
produced by GCC's config.sub.