Primary Architectures: Another Proposal (RFC)

[Kevin Kofler]

Hi,

let me suggest a new proposal for the primary architecture controversy: 
Let's have exactly one architecture handled as primary in Koji, picked with 
the ONLY aim of making builds as fast as possible. Please note that while 
this may well end up to be x86_64 at least right now, this in principle need 
not be a commonly used architecture, or even a real architecture at all, 
even something like LLVM bytecode might qualify if the gain in build time 
from avoiding the final compilation step outweights the performance loss 
from having to JIT. It might also be, say, x86_64 with some additional 
instruction set extensions (SSE3 etc.), with the actual shipped baseline 
x86_64 packages built as secondary. Anything that speeds up builds is fair 
game. Then build ALL architectures as secondary, no matter how important 
they are (e.g. at least one of the x86 architectures would be treated as 
secondary at build system level, probably both (for consistency, and to 
allow for build speed optimizations different from the packages we want to 
ship to our users)).

This proposal would only affect the build system. All the other 
differentiations between primary and secondary architectures are outside of 
its scope.

Advantages:
* faster builds (no more waiting for all architectures to complete, only the 
fastest need complete)
* fair treatment of all architectures (ARM will be happy because they won't 
be a second-class citizen anymore)

Drawbacks:
?

        Kevin Kofler