[UPDATE] Re: Update on ports on 10.0

From: Ion-Mihai Tetcu <itetcu_at_FreeBSD.org>
Date: Mon, 17 Oct 2011 15:35:51 +0300
On Tue, 11 Oct 2011 08:36:03 +0200
Erwin Lansing <erwin_at_FreeBSD.org> wrote:

> Since the release has been pushed back some more since the last mail,
> we do have some time to test a possible fix for the issues we're
> seeing with libtool on FreeBSD 10.0.  However, fixing libtool is only
> part of the problem as hundreds, if not thousands, of ports roll
> their own detection and need to be fixed individually.  We are
> currently running a fixed libtool (ports/161404) to assess how many
> ports are fixed by this patch and how many need to be patches
> manually before deciding how to move forward.  Other options include
> the big find/grep/awk solution that has been posted several times and
> fiddling with uname to go to FreeBSD 9.99 for a while, while ports
> can be fixed.
> 
> Hopefully, we can move forward in a day or two, but needless to say
> this needs a lot of testing both on 10.0 and earlier releases so we
> are sure we don't break backwards compatability, especially on 9.0
> that is soon to be released.  For those that cannot wait a few days,
> several patches have been proposed on the lists, of which dougb's
> seems most complete, so I recommend applying one of those locally.
> Please note that these are not tested widely and may break when the
> final fix is committed.
> 
> To conclude with some "fun" facts, only 232 ports break on HEAD
> currently.  Unfortunately, some of these are pretty high profile and
> prevent almost 19.000 other ports from building, leaving only slighty
> more than 3000 ports to build successfully.


Here's a little status update:
We iterated through a few -exp runs (basically for ports/161404 --
committed and ports/161431 -- skv_at_ any problem with it?). With those two
we can build around 7k packages. The majority of the rest can't be built
because of a few high profile ports that don't package: expat (6581),
curl (975), jpeg(5057), lcms(1080), libiconv(11180), libltdl(1187),
libogg(1947), pcre(5737), python27(5935).

http://pointyhat.freebsd.org/errorlogs/i386-10-latest/

What we'd like to do next is see how many ports we can package after
individually fixing those above. This will require a few other -exps
since undoubtedly we'll find other highly-depended-on ports broken that
weren't tried because of the blockers above.

Depending on this number and how long the whole process will take, we
can decide what solution to adopt. If possible we'd like to avoid the
big hammer of an uname fiddle or find/grep/sed/... (which most probably
won't work for all ports anyway, irrespective of how smart it will be).
If we need to adopt one of these hacks, it will be via some conditional
KNOB in each port Makefile, in order to have an easy way to know which
ports are fixed and which not, and an easy way to turn it off for test
builds without it in the future. Basically we do not want to shove the
dirt under the carpet, were it will rot for years.

YOU can help by sending portmgr_at_ patches for above ports (or any other
you know is broken) for the next -exp run.
And PLEASE, pretty please once you have a patch that fixes this problem
submit it upstream and bug upstream about it. (committers: please check
this is the case when committing a patch from a PR).

Thanks!


-- 
IOnut
Received on Mon Oct 17 2011 - 10:54:46 UTC

This archive was generated by hypermail 2.4.0 : Wed May 19 2021 - 11:40:19 UTC