[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: kfreebsd-* release status?



Hi,

Trying to summarise briefly, kfreebsd is ready to add the following in
jessie:

* support for KMS graphics, 3D acceleration with modern GPUs (subject to
firmware situation)

* vt(4), a UTF-8 framebuffer console (required for the above, but helps
us support more character sets too)

* a new MATE desktop replacing GNOME, XFCE still our default

* we're building more than 90% of the archive
https://buildd.debian.org/stats/graph-quarter-big.png

* XENHVM options became the default, allows to use faster PV I/O devices
than fully emulated ones

* apparently better support for KVM, Hyper-V?

* new features from OpenZFS, e.g. lz4 (faster compression)

* significant work on ZFS memory allocation, and the geom layer,
supposed to increase performance a lot

And if we get approval to put our pre-10.1 kernel in sid:

* newer KMS and vt(4) code than what we backported already

* IPSEC enabled by default

* reproducible builds of the kernel image and modules (no longer embeds
any timestamps or build paths, only what current debhelper scripts add
to the .deb)


Toolchain:  I think we're doing okay.  We moved to gcc-4.9 at the same
time as Linux arches.  Our kernel builds are changing from clang-3.3 to
clang-3.4 (but this will go most smoothly if we go with a 10.1 kernel).

With respect to openjdk-7, I've been able to keep Debian's kfreebsd
patches up to date since Februrary.  I've sent my signed copyright
assignment to Oracle so I can try to upstream them (waiting to hear back
from them).


Buildds:  fine I think

Porter boxes:  recently I heard they're slow;  DSA may be able to help
https://lists.debian.org/debian-bsd/2014/09/msg00019.html

The random thread-related hangs we saw sometimes with wheezy eglibc:
long fixed!  (Thanks to Petr's new implementation).


Onto d-i issues...

On 19/08/14 00:40, Cyril Brulebois wrote:
> [ ... let's be honest: kfreebsd-* is looking bad right
> now. ]

Thanks to KiBi for motivating me to work on d-i (and DebConf for
motivating me to generally do more Debian work these past weeks);  I
thought we had much more time but now I see why it shouldn't be left so
late to work on d-i.

Fixes are applied in unstable udebs for the following d-i issues, all of
them were regressions from d-i alpha 1 to beta 1:
#757985 - hang with ENOSPC after a few components are loaded
#757987 - cannot create swap space
#757988 - troubles with dhcp (configuration going away)

This one has been around since wheezy, is cosmetic but non-trivial to
solve;  remains an open, important issue:
#757986 - kfreebsd: multiple "inappropriate ioctl for device" prompts

> partman-* package being uninstallable. See the edos/dose reports

That happened again today (#760499 - already fixed in Git) but
surprisingly seems to have no ill effects.  In some cases
(partman-iscsi) it may even be intentional and I think we need to
distinguish those cases from others.

>> This kind of breakage is so bad that I would have expected reports
>> way earlier, or at least before my own noticing that these images
>> are unusable…

I think we got a couple of user install reports mentioning these same
bugs after KiBi had filed reports in the BTS.  We just weren't ready for
d-i beta 1.

It's understandable we get less feedback than the major linux ports,
them having 1000 times more installed machines.  But the same is likely
true of minor linux ports, like mipsel (for which I sent the only
installation-report I've seen all year?).

In kfreebsd there are a couple of things we ought to do as a result:

1. work to keep d-i usable, right up until release time, because
testers' time is valuable and we want feedback on real issues, not
about known issues - I've committed myself to this
https://lists.debian.org/debian-cd/2014/08/msg00073.html

2. automate as much testing as possible - I've shared a script and a
preseed file I've used to automate kfreebsd installs, with a view to
getting them included in jenkins.d.o
https://lists.debian.org/debian-bsd/2014/09/msg00013.html

> [...] serious bug reports in kfreebsd headers (#750836, #756553)

Fixes for those (and a regression test) pending in SVN to go into the
next upload.

> upgradability issues (kernel removal, #756464)

That remains an open, release-critical issue.


Other remaining kfreebsd RC bugs affecting jessie are *very few in
number* (OTOH I think kfreebsd began the wheezy freeze with some 50+ RC
bugs?).

#740509: ifconfig: ioctl(SIOCGIFINFO_IN6): No such device or address
That's a nasty kernel ABI break between 9.0 and 10.0, affecting
kfreebsd-i386 only.

#757711 - netcfg - has patch, I'll commit it if I get no more feedback
#711799 - grub2 - fixed in sid, waits for other RC bugs in grub2
#734451 - fuse4bsd-dkms - I suspect the package is going away for jessie
#730004 - kfreebsd-9 - won't ship with jessie
#750493 - kfreebsd-9 - won't ship with jessie

These are all the RC bugs I know of.  But there's plenty we'd still like
to fit in, such as the kfreebsd 10.1 kernel and userland.  Then lots of
polish.

Thanks!
Regards,
-- 
Steven Chamberlain
steven@pyro.eu.org

Attachment: signature.asc
Description: OpenPGP digital signature


Reply to: