Re: nasm-2.16.01,1 build (nearly) locks up machine

On Fri, 23 Dec 2022 19:50:55 +0100
Stefan Ehmann <shoesoft@gmx.net> wrote:

> On Friday, December 23, 2022 7:26:30 PM CET Tomoaki AOKI wrote:
> > On Fri, 23 Dec 2022 15:48:21 +0100
> >
> > Stefan Ehmann <shoesoft@gmx.net> wrote:
> > > On Friday, December 23, 2022 3:41:30 PM CET Yasuhiro Kimura wrote:
> > > > From: Stefan Ehmann <shoesoft@gmx.net>
> > > > Subject: nasm-2.16.01,1 build (nearly) locks up machine
> > > > Date: Fri, 23 Dec 2022 15:31:00 +0100
> > > >
> > > > > devel/nasm nearly killed my machine today:
> > > > >
> > > > > It starts thousands of gmake processes (probably until the machine
> > > > > locks
> > > > > up).
> > > > >
> > > > > I also see lots of log entries in the build log (probably ~1 per gmake
> > > > >
> > > > > invocation):
> > > > > : > asm/warnings.time
> > > > >
> > > > > gmake asm/warnings.c.time include/warnings.h.time
> > > > > doc/warnings.src.time
> > > > > gmake[13499]: Entering directory '/wrkdirs/usr/ports/devel/nasm/work/
> > > > > nasm-2.16.01'
> > > > >
> > > > > Seems some kind of make loop. Is this a general problem or is there
> > > > > something wrong with my setup? I'm building in poudriere on
> > > > > 13.1/amd64.
> > > >
> > > > Do you build nasm 2.16.01 with poudriere and you tmpfs for working
> > > > directory (that is, you specify either 'USE_TMPFS=wrkdir' or
> > > > 'USE_TMPFS=yes' in poudriere.conf)?
> > > >
> > > > If so, you can work it around by disabling use of tmpfs by adding
> > > > 'USE_TMPFS=no' in poudriere.conf.
> > >
> > > Yea, I have USE_TMPFS=all in poudriere.conf. USE_TMPFS=no works indeed.
> > >
> > > Thanks for the fast reply.
> >
> > I've bitten by it, too.
> > Found this. [1] [2]
> > [2] is linked from [1].
> >
> > But Not at all tried though (as I'm not sure it is safe [100% upper
> > compatible] to switch to -devel).
> >
> > ports-mgmt/poudriere-devel seems to have it, but ports-mgmt/poudriere
> > doesn't.
> >
> > [1] https://github.com/freebsd/poudriere/issues/888
> >
> > [2]
> > https://github.com/freebsd/poudriere/commit/56233a1aaea1be59dcc111e7b3f97b6e
> > 891bb06a
> 
> Thanks for the pointer. The rust tmpfs problem is different - it will work if
> you have enough RAM. For some reason (not obvious to me), nasm build is
> totally broken on tmpfs.
> 
> Meanwhile, the nasm update was reverted in commit
> 102f7173426ac4122afd355205aa141bb50b2803.

The most important thing is that [2] introduced TMPFS_BLACKLIST
variable.

I think adding nasm there causes building nasm without using tmpfs,
while other ports, excluding listed there at the same time, can keep on
using tmpfs.

P.S.
To test pre-built nasm causes failure on ports BUILD_DEPENDing on nasm,
I've done below and confirmed OK, even with USE_TMPFS=yes
on /usr/local/etc/poudriere.conf.
So maybe TMPFS_BLACKLIST'ing nasm alone would be sufficient.

 1. Built devel/nasm with ports-mgmt/pkg_replace, which builds fine,
 2. pkg create it,
 3. start usual poudriere run with unusual "-i" added,
 4. after poudriere goes into shell, copy pre-built nasm package,
    to /poudriere/data/packages/(jain name)/All/,
 5. exit shell invoked by poudriere,
 6. once finished, delete jpeg-turbo package from
    /poudriere/data/packages/(jain name)/All/,
 7. build jpeg-turbo using poudriere.

Copying pre-build nasm package without steps 3-5 caused re-building
nasm, which promissingly fails.

-- 
Tomoaki AOKI    <junchoon@dec.sakura.ne.jp>