variant or platform

[Weissmann Markus]

On 15 Jul 2007, at 23:53, Anders F Björklund wrote:

> Ryan Schmidt wrote:
>
>>> Seems like all the current ports with "x86" or "intel" as the  
>>> arch are
>>> either wrong or typos, though... "darwin x86" should be "darwin  
>>> i386",
>>
>> I still say that this needs to be examined and evaluated  
>> individually for each port. If nobody's complaining about these  
>> ports, then maybe whatever's being done in the darwin x86 platform  
>> selector isn't necessary for Intel Macs at all, and might in fact  
>> be harmful.
>
> It's just 7 ports, but so far it all looks like they should be  
> changed.
>
>> For example, mpfr has a section "platform darwin x86", but it also  
>> has an identical section "platform darwin i386", so the port  
>> maintainer, Vincent, should be asked why this is duplicated. I'm  
>> Cc'ing him on this email.
>
> I have no idea what kind of Darwin system would return "x86" in uname.
> The list at http://wiki.tcl.tk/1649 only mentions "i386" and "i386"...
>

OpenDarwin did return x86 while Mac OS X returns i386; so the 'x86'  
stuff is/was for OpenDarwin while 'i386' is the Mac OS X Intel stuff.
So please do NOT replace x86 by i386 by default - this is not an  
error to be corrected this way. If no one complains, we may just  
remove the x86 ones as there is no OpenDarwin distribution anymore.


>> And xgalaga has a section "platform darwin x86" which should  
>> clearly be "platform darwin i386", and I've changed that now, but  
>> the port doesn't build on Intel Macs anyway because it uses (and  
>> seems maybe to require) gcc 3.3.
>
> See other email... There is no such i386 gcc-3.3 for Mac OS X 10.4 ?
>
>>> and the one (1) port with "darwin intel" is just plain wrong IMHO.
>>
>> Ah -- "darwin intel" -- that's why I didn't find it before. Got it  
>> now: it's libdnsres. I'm Cc'ing the maintainer, Mark, since he  
>> should have a say in this, but I agree, this should be "platform  
>> darwin i386". I think he should also remove the unnecessary CFLAGS  
>> and LDFLAGS definitions.
>
> Actually it should be "platform darwin 8 i386", but most likely it
> should be updating the config.sub and config.guess or so instead ?
>
>>> They should most likely be *standardized* to be "powerpc" or "i386",
>>> even though the aliases "ppc" and "x86" are theoretically valid too
>>> (for example Darwin 6-7 was i386, Darwin 8-9 are i686, both: "x86")
>>
>> I don't know what you mean by all of this.
>
> The triplets are:
> i386-apple-darwin6
> i386-apple-darwin7
> i686-apple-darwin8
> i686-apple-darwin9
>
> Other operating systems, like Fedora, went through similar changes  
> as well.
> (whether arch actually returns i386 or i686 depends on some other  
> factors)
>
>> I'm on darwin 8 now, and uname returns i386, and "platform x86"  
>> sections definitely do not get executed by MacPorts on an Intel  
>> Mac currently. If you're suggesting that MacPorts base could be  
>> changed so that x86 would also work on Intel Macs, then I again  
>> reiterate that I believe x86 is being used in several ports right  
>> now in an attempt to target non-Macs. If such a change were made  
>> to base, those ports would start failing on Intel Macs.
>
> I meant that i386/i686/x86 are all the same thing, and "i386" is  
> the name.
>
> Optimizations for certain processors are done elsewhere anyway,  
> here it
> is just talking about the generic architecture - Intel x86 or  
> PowerPC ?
>
>>> It's easy to tweak port.tcl to return "i386" for any "x86" or "i686"
>>> machines from tcl_platform, just as it is currently returns  
>>> "powerpc"
>>> for the hopelessly silly `uname -m` return value of "Power  
>>> Macintosh"...
>>
>> It might have been clearer if we had tweaked "i386", "i686", and  
>> "x86" to be just "intel", but I'm not sure if we want to go to all  
>> the trouble to do that now.
>
> I prefer the shorter x86 and ppc myself, but it doesn't matter much.
>
> The current arch names of "i386" and "powerpc" works just as well,
> but if you want to introduce the extra alias "intel" for i386 then  
> do...
>

If you want this done 100% correctly, I'd recommend "ia32" (Intel  
Architecture 32), "aa64" (AMD Architecture 64), "ppc32" and "ppc64" -  
and of course 'arm32'... ;)


>>> The suggested "port platform" info command could help people  
>>> determine
>>> what values to put into their platform variants, but for the current
>>> common target it's probably enough to make a list of available  
>>> ones ?
>>
>> I'm not sure what you mean here? What suggested "port platform"  
>> info command?
>
> I formalized the discussion we had earlier into a Trac ticket, #12302
>
> Basically it returns something that would work if put as platform {}
>
> --anders
>
> _______________________________________________
> macports-dev mailing list
> macports-dev at lists.macosforge.org
> http://lists.macosforge.org/mailman/listinfo/macports-dev

---
Markus W. Weissmann
http://www.mweissmann.de/