Benefits of latest gas?

[Stan Shebs]

Ian Lance Taylor wrote:

>Stan Shebs <shebs@apple.com> writes:
>
>
>>As some of you may know, Mac OS X has long been using gas 1.38, pre-BFD
>>and all, as its system assembler, heavily hacked to do Mach-O and
>>support all of Apple's various extensions. Recently I had occasion
>>to look at the some of the barbaric expression code in our version,
>>and thinking "surely the latest code is much improved", went and looked
>>at the latest version, and it didn't really seem changed by that much,
>>which was disappointing. In fact, quite a bit of 1.38 seems the same as
>>current gas sources.
>>
>>Up to now I've been generally assuming that staying on the old gas was
>>a poor strategy, and that the massive effort that would be needed to get
>>Mach-O into latest gas would be an investment quickly repaid in better
>>quality, better error-checking, etc. But now I'm not so sure. We really
>>only have the one target (PPC), plus x86 for Darwin, so portability
>>arguments aren't that compelling, and the assembler is pretty much in
>>the noise compile-time-wise already.
>>
>>So what are the features of latest gas vs 1.38 that would make my
>>management whip out the checkbook for a merge/upgrade project?
>>
>
>I would say that the major new features are for people who program in
>assembler: macros, better conditionals, better error handling, more
>pseudo-ops.  For C/C++ code, I would say that as long as you stick
>with Mach-O, you aren't going to see much improvement by upgrading to
>the current assembler.  I don't even know whether the current
>assembler is faster--the pre-2.0 versions were pretty fast.
>
>Over time you might some savings as other people do work for
>you--adding support for new processor variants, fixing bugs, speeding
>up the assembler, etc.
>
>
Thanks, that's some good insight!

>But all in all I think you would be better off using that checkbook to
>pay your kernel hackers to add ELF support.  That would get you the
>current gas for free, and I think it would simplify your life quite a
>bit over the long run.
>
It's been suggested a few times. :-) Mach-O has had a lot of special
features added over the years - unusual kinds of directives and relocs,
multi-architecture files, etc, all of which OS X development depends
on. Getting the equivalent functionality into ELF would not be
impossible, but certainly quite expensive. The kernel work is very
nearly the easiest part of the project.

Stan