What we have here, ...

[Warner Losh]

On Jun 14, 2014, at 9:11 PM, Glen Barber <gjb at FreeBSD.org> wrote:

> On Sat, Jun 14, 2014 at 09:03:51PM -0600, Warner Losh wrote:
>> 
>> On Jun 14, 2014, at 8:51 PM, Glen Barber <gjb at FreeBSD.org> wrote:
>> 
>>> Is a failure to build head/ with high make(1) -j values.  Again.
>> 
>> There’s likely a dozen or more of these lurking in the tree. Ian’s last set
>> of patches squashed many of the ones in lib, but I’ve not done a careful
>> audit of the i18n code.
>> 
> 
> Ok, thanks for the info.  I'm a bit unclear on how the parallelization
> stuff in share/ actually works, but I'm happy to help crowbar at things.

I haven’t even looked, to be honest...

>>> With empty /usr/obj, I have been seeing strange build failures on head/
>>> with high -jN values, but even as low as -j10.  The machine used to
>>> build snapshots of head/ and stable/ branches has been using -j48 for
>>> several months without issue, so I am certain this is a recent (<=1
>>> month) issue.
>>> 
>>> Of all things, it looks like fonts are failing to build, giving the
>>> following error:
>>> 
>>> --- MACGUJARATI.esdb ---
>>> NAME is mandatory.
>>> ENCODING is mandatory.
>>> *** [MACGUJARATI.esdb] Error code 1
>>> make[6]: stopped in /usr/src/share/i18n/esdb/APPLE
>>> 
>>> Is anyone else seeing this with empty MAKEOBJDIR ?
>> 
>> I haven’t seen it, but in -j races, that means approximately zilch… 
>> 
> 
> Yeah.  To be honest, it's the similar situation I emailed you privately
> about a few weeks ago with the race in rescue/ that makes no sense.  So
> while I don't entirely trust make(1) reporting what blew up, unlike the
> issue in rescue/, the 'NAME is mandatory' error has been about 90%
> reproducible between both amd64 and i386.  (Without being able to build
> at least amd64, I cannot test other architectures.  Useless data point,
> I know.)

Yea, that turns out to be a cascade of dozens of errors based on a file being
empty…. The error reporting is correct, you just have to thread it back a ways.
Sure would be nice to have a tool that would remove the clutter from the make
that we’ve grown when you’re looking for a specific thing.

>>> I won't be able to reproduce the issue with log output until at least
>>> Monday, since I'm running the head/ builds with the '-jN' halved to 24,
>>> which so far seems to have made a difference.
>> 
>> How many cores in this box of yours?
>> 
> 
> 48.

I only have 12 (or 24 if I configure for threading) :).  That might make a big difference.

Warner
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 842 bytes
Desc: Message signed with OpenPGP using GPGMail
URL: <http://lists.freebsd.org/pipermail/freebsd-current/attachments/20140615/4e3b12f4/attachment.sig>