traditional mips vs. little endian?

[Andrew Cagney]

> I was thinking changing the default abi, but your arguments are
> compelling...


I'm guessing your meaning for things like mips-elf and the like.


>> 
>> * making the test-case consistent w/ the configury as it is now, or
>> 
>> * Making only IRIX use the SGI ABI ('non-traditional'), and move all
>> existing non-os-specific 'embedded' targets (incl. mips-elf and
>> mipsel-elf) over to 'traditional mips' (or worse, even the non-IRIX
>> OS-specific ones)?
>> 
>> 
>> I don't really like latter choice, but don't _deeply_ care either.
>> Certainly it would make things more consistent, but is it really
>> desirable to change the ABI and configuration in use for ... who knows
>> how long, for those targets?
> 
> 
> Good point.  I'd like to use a different ABI by default, but I'm not
> sure what the deal would be with breaking all of the previous.  Does
> anyone have any advice here?


This feels a bit like replacing history with flavor of the month. 
Remember that changing the ABI/tuple/... and how they are all 
interpreted affects things starting at the compiler and going right 
through to the debugger.

More importantly, it affects the users (well my type of user :-) and 
their expectations.  I don't think users are going to be happy if things 
  as entrenched as an ABI, for their target, changes between each release.

Perhaphs the question to ask is: is this being done for technical or 
asthetic reasons?

Glad it isn't my decision.

	enjoy,
		Andrew