BuildRoot and mktemp -d

[Ville Skyttä]

On Tuesday 13 February 2007, you wrote:
>  The following page has been changed by TomCallaway:
>  http://fedoraproject.org/wiki/Packaging/GuidelinesTodo 
[...]
>  + ||ratify ||New Required Buildroot||[wiki:Self:TomCallaway
>  spot]||2007-02-13||Change required buildroot for new packages to %(mktemp
>  -d %{_tmppath}/%{name}-%{version}-%{release}-XXXXXX) (old packages/core
>  merge packages using old buildroot are grandfathered)|| 
Using mktemp -d in specfiles' BuildRoot means that quite a few stray temp dirs 
will start to appear in %{_tmppath}.  For example "rpm -q --specfile 
foo.spec" and "rpmbuild -bs foo.spec" create them, and nothing cleans them
up 
(no, tmpwatch doesn't count) - we probably don't want that.

One way to avoid it is mktemp -ud, but it's more racy than plain -d.  I 
think -ud would be a better choice nevertheless.

[R P Herrold]

On Wed, 14 Feb 2007, Ville Skyttä wrote:

>  Using mktemp -d in specfiles' BuildRoot means that quite a 
>  few stray temp dirs will start to appear in %{_tmppath}. 
>  For example "rpm -q --specfile foo.spec" and "rpmbuild -bs 
>  foo.spec" create them, and nothing cleans them up (no, 
>  tmpwatch doesn't count) - we probably don't want that. 
umm, if it is to be so done,  passing mktemp a subordinate 
'template' directory pattern of

 	-p %{_tmppath}rpm-build/{...whatever nameing per build instance}

may be helpful in confining this to a known search sub-tree in 
%{_tmppath}; I know I go poking in the RPM-BUILD temp dirs on 
occasion when builds are being diagnosed, but also like to 
clean house confidently when space gets tight.

-- Russ Herrold

[Rex Dieter]

Ville Skyttä wrote:

>  Using mktemp -d in specfiles' BuildRoot means that quite a few
> stray temp dirs 
>  will start to appear in %{_tmppath}.  For example "rpm -q --specfile 
>  foo.spec" and "rpmbuild -bs foo.spec" create them, and nothing cleans them
> up 
>  (no, tmpwatch doesn't count) - we probably don't want that. 
Isn't that what %clean is for?

>  One way to avoid it is mktemp -ud, but it's more racy than plain
> -d. 
I doubt this would work.

-- Rex

[Jon Masters]

Rex Dieter wrote:
>  Ville Skyttä wrote:
>  
> > Using mktemp -d in specfiles' BuildRoot means that quite a few stray temp dirs 
> > will start to appear in %{_tmppath}.  For example "rpm -q --specfile 
> > foo.spec" and "rpmbuild -bs foo.spec" create them, and nothing cleans
> them up 
> > (no, tmpwatch doesn't count) - we probably don't want that.
>  
>  Isn't that what %clean is for? 
That's what I use it for...

Jon.

[Jason L Tibbitts III]

> >>>> "RD" == Rex Dieter
> <rdieter(a)math.unl.edu&gt; writes: 
RD> Isn't that what %clean is for?

Not really; the problem here is that even parsing of the spec will
call mktemp and thus create some random directory.

This kind of thing needs to be in deeply coded into rpm, and until
then we just need to stick with the single approved buildroot as we
have now.

 - J<

[Ville Skyttä]

On Wednesday 14 February 2007, Rex Dieter wrote:
>  Ville Skyttä wrote:
>  > Using mktemp -d in specfiles' BuildRoot means that quite a few stray temp
>  > dirs will start to appear in %{_tmppath}.  For example "rpm -q --specfile
>  > foo.spec" and "rpmbuild -bs foo.spec" create them, and nothing
> cleans
>  > them up (no, tmpwatch doesn't count) - we probably don't want that.
>
>  Isn't that what %clean is for? 
%clean is not run with eg. "rpm -q --specfile" nor "rpm -bs".  And
it's not 
even possible to pass --clean to the former, and passing it to the latter 
does not clean up the created temp dir either - it doesn't run the 
actual %clean scriptlet from the specfile.

>  > One way to avoid it is mktemp -ud, but it's more racy than
> plain -d.
>
>  I doubt this would work. 
Why not?  Seems to work fine here in the few tests I've done.

[Rex Dieter]

Ville Skyttä wrote:
>  On Wednesday 14 February 2007, Rex Dieter wrote: 
> >> One way to avoid it is mktemp -ud, but it's more racy
> than plain -d.
> > I doubt this would work.
>  
>  Why not?  Seems to work fine here in the few tests I've done. 
Fair 'nuf, as long as it works. (:

-- Rex

[Axel Thimm]

On Wed, Feb 14, 2007 at 12:04:39AM +0200, Ville Skyttä wrote:
>  On Tuesday 13 February 2007, you wrote:
>  >
>  > The following page has been changed by TomCallaway:
>  > http://fedoraproject.org/wiki/Packaging/GuidelinesTodo
>  [...]
>  > + ||ratify ||New Required Buildroot||[wiki:Self:TomCallaway
>  > spot]||2007-02-13||Change required buildroot for new packages to %(mktemp
>  > -d %{_tmppath}/%{name}-%{version}-%{release}-XXXXXX) (old packages/core
>  > merge packages using old buildroot are grandfathered)||
>  
>  Using mktemp -d in specfiles' BuildRoot means that quite a few stray temp dirs 
>  will start to appear in %{_tmppath}.  For example "rpm -q --specfile 
>  foo.spec" and "rpmbuild -bs foo.spec" create them, and nothing cleans them
> up 
>  (no, tmpwatch doesn't count) - we probably don't want that.
>  
>  One way to avoid it is mktemp -ud, but it's more racy than plain -d.  I 
>  think -ud would be a better choice nevertheless. 
Good catch! +1
-- 
Axel.Thimm at ATrpms.net