Re: *** TRYING TO INSTALL GUIXSD v0.15.0 FOR DAYS ***

[Mark H Weaver <mhw@netris.org>]

Hi Brian,

Brian Woodcox <bw@inskydata.com> writes:

> This is what is displaced on the screen when the hang occurs:
>
> …
> make  check-TESTS
> make[3]: Entering directory ‘/tmp/guix-build-guile-2.2.3.drv-0/guile-2.2.3’
> Testing /tmp/guix-build-guile-2.2.3.drv-0/guile-2.2.3/meta/guile …
> with GUILE_LOAD_PATH=/tmp/guix-guild-guile-2.2.3.drv-0/guile-2.2.3/test-suite

Thanks.  These last messages were printed by the 'check-guile' script,
just before it passes control to test-suite/guile-test, which runs the
tests in test-suite/tests/*.test.  I guess it's getting stuck during the
initialization of 'guile-test', because it prints a message before
running each test.

At this point, I see a few possible next steps, from easy to harder:

(1) You could wait until hydra.gnu.org comes back online, which I expect
to happen sometime next week.  Hydra has a full set of substitutes for
0.15.0, so you shouldn't need to build Guile at that point.

(2) You could try QEMU instead.  I suspect that Guix has seen far more
testing under QEMU than VirtualBox, because QEMU is in Guix and not
VirtualBox.  If it fails in QEMU, then we will have a test case that
Guix developers can try to reproduce on their own systems.

(3) If you felt like getting your hands dirty and digging deeper to
investigate this problem and find its source, read on:

You could try the same build with "--keep-failed" added to the Guix
command line, and interrupt it after it gets stuck.  At that point, you
should have write access to /tmp/guix-guild-guile-2.2.3.drv-0, and you
can enter that directory and try various experiments.

/tmp/guix-guild-guile-2.2.3.drv-0/environment-variables will contain the
environment variable settings that were passed to the top-level build
commands, including "make check".  In that directory, run:

  env -i `which bash`

or a similar command to clear the environment, and then "source
environment-variables" to load the environment settings.  Then 'cd' into
the guile source directory and run ./check-guile.

Hopefully it will get stuck here as well.  If so, you could use GDB to
attach to the stuck guile process and obtain a backtrace.  It might also
be useful to see the output of "strace -f ./check-guile".

There are many other things that could be tried at this point, such as
removing the "--debug" flag passed to guile at the end of 'check-guile',
inserting debugging print statements at various points in the
'guile-test' script, etc.

Hopefully one of these options is workable.  Sorry for the bother.

     Regards,
       Mark