On Fri, Mar 05, 2004 at 10:07:42PM +0900, GOTO Masanori wrote: > > > > That's crap. m68k is a lot more slow than most SH machines, and I am > > (mostly) keeping up with sid on my m68k mac; also, there is no real > > problem building sid for m68k, because we have 10 autobuilders or so for > > m68k (a bit less now, due to some more or less political issues). This > > mail is written at the console of my m68k mac, BTW :-) > > > > It's most certainly possible; whether the demand is there, that's > > something else. > > I think the reason is simple. Debian developers don't have rich SH-3 > or SH-4 hardware (ex: memory). It's neither ABI issue nor speed. Is it possible to set up cross-building buildd's instead? The fastest SH machine I know of is a Dreamcast, and that is still slow (200Mhz, 16MB). It would seem that if we could build the SH distribution from e.g. a fast i386 machine, it would be more manageable. Does anyone have a comprehensive list of SH-3 and SH-4 based machines that would be capable of running Debian? Is any machine faster than the Dreamcast? Regarding the ABI issue, I think there is a fundamental conflict that is difficult to solve. If we define a new ABI for SH to combine SH-3 and SH-4 ABI, then we have become incompatible with other existing object code. (Is this really important?) It would also take expertise and some time to define a new ABI. If we do not define a new ABI, we place the load of another separate distribution on the mirrors' disk space and bandwidth, which doesn't even really have that many users. This would be unacceptable to the Project I think, because we would be using a disproportionately large amount of resources to serve a very small userbase (arguably the smallest user base out of all Debian architectures). Why are the SH-3 and SH-4 ABI different in the first place? This seems stupid for Hitachi to do such a thing when the hardware is so similar. -- Ryan Underwood, <nemesis@icequake.net>
Attachment:
signature.asc
Description: Digital signature