Re: guix build freezes

[Ludovic Courtès]

Ricardo Wurmus <address@hidden> skribis:

> I have a very weird problem that I don't understand.  "./pre-inst-env
> guix build" just sits there doing nothing at all:
>
>     address@hidden:/localhome/rwurmus-tmp/guix] (531) $ git describe HEAD
>     v0.8.1-1597-g4702cec
>     address@hidden:/localhome/rwurmus-tmp/guix] (532) $ time ./pre-inst-env 
> guix build texlive-texmf-2014
>     The following file will be downloaded:
>        /gnu/store/h1qjkg8bvcdb485q81lw5844szgywpg7-texlive-texmf-2014
>     ^C
>
>     real      13m0.888s
>     user      0m3.728s
>     sys       0m0.364s
>
> Now using the previously "sudo make install"'d version of Guix:
>
>     address@hidden:/localhome/rwurmus-tmp/guix] (533) $ time guix build 
> texlive-texmf-2014
>     guix build: warning: failed to load '(gnu packages ld-wrapper)':
>     ERROR: no code for module (gnu packages ld-wrapper)

(Side note: there’s a stale ld-wrapper.scm file here, hence the warning.
The problem with repeated ‘make install’ is that files are left behind.)

>     /gnu/store/0zx4rfbf58s0fnh4nzcaajdrfwb1ilv5-texlive-texmf-2014
>
>     real      0m24.060s
>     user      0m4.004s
>     sys       0m0.541s
>
> I can't really say when I last make install'd, but it was probably
> around commit 27350ef8b.

Uh, I can’t reproduce it.  Could you try:

  ./pre-inst-env strace -o log guix build texlive-texmf

to see where it’s stuck?

Another issue: 24 seconds is really a *lot*, esp. if there was nothing
to build.  Where did those 20 seconds between user and real time go?
Could it be due to a very slow file system?  Is the daemon socket file
stored on NFS?  (Does that even work?)

> I have a feeling I should do "make clean" again and purge everything,
> but I thought I should report this anyway.  I should also note that I
> develop Guix on a different machine with the *same* network-mounted home
> directory.  On guix-builder I make and make install everything from
> within a directory on a local disk, but my actual home directory is the
> network share, so I suspect that this could all be explained by sharing
> some go caches or so.  Could someone please confirm that this is in fact
> what's happening?

The .go files are loaded either from $prefix or from $builddir (when
using ./pre-inst-env.)  If these are on NFS, that certainly slows things
down, but I don’t think there are other side effects.

HTH,
Ludo’.