On Fri, Jan 14, 2005 at 02:47:01AM -0500, Joey Hess wrote: > Christian Perrier wrote: > > I'm not even sure that hppa installs are really well tested, by the > > way. Joey could tell more and this probably needs more input from him. > > He may even correct me if I'm completely wrong. > So all I can say for sure about d-i 2.6 is that it works for netboot on > one model of machine for basic installs. I don't have much depth or > breadth of information. But then, we don't have much more information > about how well 2.4 works on hppa. What we do know is that it pretty much > sucks, at least 50% of installation reports show it failing. > Freezes in the middle of debootstrap in CD installs with 2.4 have been > reported quite a lot for the last 10 months. > But really the available info is slim. Some recent reports of failure > include #287606 (which has 2.6 failing too), #282532, #271011, and > #269278 (two people seeing the CD copy freeze), #265420, #264433 (hang > at boot from CD). That takes us back to August. In the same time period > I see more or less successful reports #263207, and #283754 (mine). Um, > that's all. I'll trust the hppa kernel guys if they think 2.6 is likely > to improve these stats. > What I'm most fearful of is that we don't know if d-i works at all on > hppa for CD installs with 2.6. As far as I know the closest thing to CD > install ever was a few partial installs bdale did that way while working > on hppa 2.6 support in d-i. There could be any number of problems that > we've yet to find in getting that sorted out and really working; I'd be > much more comfortable with this whole thing if we had 2.6 hppa cd images > (it has to be a separate image afaik due to palo's inflexability), and > if we knew they worked at least as well as netboot works. > Given how few installation reports we get about hppa at all, I don't > share vorlon's caution as far as not wanting to switch to 2.6 as default > or drop 2.4 until it's been fully tested in a d-i release. But I'm going > to have to see it work, and seem to work on as many machines as 2.4 > before I can support it. If the existing 2.4 kernels for hppa are consistently in this bad of shape (I only knew that they were slow, not broken), then as much as it hurts my stomach, I'm less opposed to trying to make this switch in RC3 if it's agreed that this is the thing to do. I do also agree that this still means we need to have successful CD installs before RC3 if we're expected to support CD as an install medium for hppa in sarge. Still, if the patch job to get 2.4.27-2 in the archive for hppa is this big a deal, we would need to get new install reports *anyway* to make sure the new kernels aren't hosed, and getting hppa d-i ported to 2.6 is a better use of developer time than backporting patches to 2.4, sure. > > Well, again, Joey Hess could probably tell more but even if what you > > mention above has been explained, not having enough interaction > > between the hppa kernel team and the d-i team is quite likely to be > > the real problem here. > It's been no secret that the 2.4 kernel was not in good shape and hppa > folk were happier with 2.6. But this idea of just dropping 2.4 > altogether was a suprise to me. > > Just telling "well, guys, we want to abandon 2.4 ASAP" is probably not > > enough and is quite likely to be either ignored or missed if noone > > motivated jumps in debian-boot and discusses this issue more > > deeply. After all, not everyone (and maybe noone) in the d-i team is > > thinking daily about hppa installs... > We have maybe two people, each of whom are more involved in 3 or 4 other > arches. Not enough attention to get hppa 2.6 support fully done in time > for rc2 even though it was one of a few things I delayed that release > over for 2 or 3 weeks. Thibaut, I hope not having to merge hppa patches for 2.4.27-2 means you and other pa-risc kernel folks might have some time to help verify that the installer works with 2.6, and help get it working if it isn't already? Thanks, -- Steve Langasek postmodern programmer
Attachment:
signature.asc
Description: Digital signature