Re: TMPDIR and bsd.pkg.mk (was Re: clang builds and /var/tmp)

[Richard PALO <richard.palo%free.fr@localhost>]

Le 28/10/15 20:13, Martin Husemann a écrit :
> On Wed, Oct 28, 2015 at 02:35:56PM -0400, Greg Troxel wrote:
>> I haven't had time to really think enough about this, but I tend to
>> agree with Joerg.    I think the real problem is that TMPDIR should
>> essentially never be /var/tmp for building.   It would seem reasonable
>> to think about forcing TMPDIR=/tmp in the environment in the various
>> opsys.mk files, essentially fixing bugs in the underlying OS that lead
>> to compiler intermediaries being in a real fs vs ram fs.
> 
> I have lots of tiny boxes that I build native packages on "once in a
> while" that have a very limited tmpfs on /tmp and *need* to use
> /var/tmp for anything bigger.
> 
> Martin
> 

Hi Martin, in this environment, you probably set WRKOBJDIR to something
under /var/tmp or leave it in tree.

What I'm currently testing in bsd.pkg.mk:
>> -ALL_ENV+=              HOME=${FAKEHOMEDIR}
>> +ALL_ENV+=              HOME=${FAKEHOMEDIR} TMPDIR=${FAKEHOMEDIR}
will simply keep things there as well.

What has not been dealt with [yet] is if it is eventually appropriate to
push that elsewhere... for example something like user or pkg defined PKGTMPDIR.
In which case, TMPDIR could be something like TMPDIR=${PKGTMPDIR:U${FAKEHOMEDIR}}

Myself I keep WRKOBJDIR in '/tmp/pkgsrc' and have only come across so far an issue when building clang as the build directory pushes over 24G!  Naturally using /var/tmp/pkgsrc
easily gets over this, but that means /var/tmp must also be able to handle the size.

I'm in absolutely no hurry to rush the bsd.pkg.mk patch, which will certainly need more experience "in the field".

-- 
Richard PALO