ARM as a primary architecture

Kevin Kofler kevin.kofler at
Thu Mar 22 03:06:26 UTC 2012

Tom Lane wrote:
> To put it as succinctly as possible: monocultures are bad.

That's what secondary architectures are for.

> Do you really think that x86 will be the most desirable architecture
> forever?  Things change fast in this business, and that arch is weighted
> down by enough bad ancient decisions that I think it's eventually going to
> lose out.

That's when it will become the time to promote the secondary architecture 
that eventually wins out to primary (and to start working on a plan to phase 
out x86 and demote it to secondary as we did with PPC). Not now. We don't 
even know whether ARM will be the architecture that eventually wins out yet!

> I thought it was a serious error to drop PPC from primary-arch status.

I think it was one of the best decisions Fedora ever made. I'm glad I don't 
have to deal with slow PPC builders anymore, nor to fix build errors for 
such an obsolescent architecture. PPC stopped being relevant the day Apple 
switched to x86. The GNU/Linux on PS3 efforts managed to delay its death by 
a couple years, but in hindsight that kind of platform is really what 
secondary architectures are for! (It's not even a computer.) I think PPC had 
actually been primary for way too long, the demotion to secondary was long 
overdue when it was finally implemented.

> But now that we've done that, putting in another one should be a high
> priority wish-list item.

I strongly disagree on that point. Non-x86 primary architectures are a major 
pain that really needs to be avoided. Doing it just for the sake of having a 
non-x86 primary architecture is just completely wrong! I really don't miss 
the fights I had with ppc/ppc64 as a primary architecture (it was a royal 
PITA for KDE packaging, some KDE module or dependency kept breaking due to 
platform-specific toolchain bugs or limitations (ever heard of the infamous 
PPC TOC? There's not much you can do when autogenerated SWIG bindings exceed 
the PPC64 TOC size, and minimal-toc doesn't help when it's all in one 
translation unit), and most of our builds were glacially slow due to PPC) 
and am most definitely NOT looking forward to seeing this again.

> I'm as concerned as anyone about whether we can (in the near future) get
> ARM builders that are fast enough to make it *practical* for ARM to be a
> PA.

Me too! (At least one thing we agree on. ;-) )

> But I think denying that we need non-Intel PAs is just fundamentally
> wrongheaded.

And I couldn't disagree more (for the reasons explained above).

        Kevin Kofler

More information about the devel mailing list