On Jan 29, 2013, at 8:55 PM, Gordon Ross <[email protected]> wrote:
> On Mon, Jan 28, 2013 at 11:06 AM, Garrett D'Amore <[email protected]> wrote: >> I'm sorry to open a thread that I'm sure will cause a lot of discussion most >> of which will probably be fruitless and pointless. And yet, I feel >> compelled to do this. Please think before you respond, and only reply if >> you have some thing useful to add to this conversation. >> >> Recently, I advocated (as in RTI advocacy), a change to a program - kstat. >> After I integrated this change, another advocate pointed out that the SPARC >> port was broken by this otherwise worthwhile change. In that particular >> instance, the advocate took it upon himself to fix the problem. >> >> However, its clear to me that this particular advocate (Richard Lowe) may be >> the only advocate with regular access to SPARC. Its also clear to me that >> the vast majority of ordinary developers lack such access to SPARC. > > In thinking about this, a question comes to mind? > What did this "SPARC breakage event" cost you? This time, not much. It did cost Rich Lowe some time, and me some overall aggravation. > > My impression is that keeping SPARC around has not cost us much. > Sure, it gets broken more often than intel, and it stays broken until > someone with time and interested comes along to fix it. So? Actually, up until now, we've had a policy that we backout or fix changes that bust SPARC. We came very close to doing that this time -- Rich Lowe's effort to address it is what prevented the backout. And to be clear, someone had to do real engineering work to fix a problem on SPARC; so it was not free. Maybe not very expensive -- but definitely not free. > > My suggestion would be to just relax a little, rather than treat each > SPARC breakage event as some emergency. We can give all > these SPARC fans who eagerly respond to your thread a chance > to come help keep SPARC alive if they want. I hope they will. Ick. If we are going to kill support for SPARC, lets be honest about it. The idea that we maintain a bunch of code in our tree that is basically broken, and we know it, is unacceptable to me. Its also not acceptable to impose a burden on just one or two individuals who do care that we don't have broken code in our tree. I'd rather wholesale remove SPARC entirely (and let someone else fork the tree), than let it fall down the quality death spiral. I'm *strongly* of the opinion that we need to either have it as a first class port, or eliminate it. And in case anyone thinks I want SPARC to go away -- you are very much mistaken. I would rather actually have SPARC in tree -- I have both fond memories of this platform, and I think there is real value in having a viable big-endian architecture in our tree -- I believe our *quality* is better for SPARC. (I *know* our device drivers are better thanks to the fact that SPARC requires a little extra care than x86, this comes partly from endianness -- but more so from the IOMMU and the additional requirements that this imposes.) But if we can't keep the resources running to keep this a viable first class port, then we should just kill it. - Garrett ------------------------------------------- illumos-discuss Archives: https://www.listbox.com/member/archive/182180/=now RSS Feed: https://www.listbox.com/member/archive/rss/182180/21175430-2e6923be Modify Your Subscription: https://www.listbox.com/member/?member_id=21175430&id_secret=21175430-6a77cda4 Powered by Listbox: http://www.listbox.com
