bug#29368: Unreliable failing tests / segfaulting guile

[Martin Castillo]

Hi,

On 24.11.2017 18:02, Ludovic Courtès wrote:
> Hi,
> 
> address@hidden (Ludovic Courtès) skribis:
> 
>> Martin Castillo <address@hidden> skribis:
>>
>>> test-name: derivation-prerequisites-to-build when outputs already present
>>> location: /home/mcd/guix/tests/derivations.scm:790
>>> source:
>>> + (test-assert
>>> +   "derivation-prerequisites-to-build when outputs already present"
>>> +   (let* ((builder '(begin (mkdir %output) #t))
>>> +          (input-drv
>>> +            (build-expression->derivation
>>> +              %store
>>> +              "input"
>>> +              builder))
>>> +          (input-path
>>> +            (derivation-output-path
>>> +              (assoc-ref (derivation-outputs input-drv) "out")))
>>> +          (drv (build-expression->derivation
>>> +                 %store
>>> +                 "something"
>>> +                 builder
>>> +                 #:inputs
>>> +                 `(("i" ,input-drv))))
>>> +          (output (derivation->output-path drv)))
>>> +     (when (valid-path? %store input-path)
>>> +           (delete-paths %store (list input-path)))
>>> +     (when (valid-path? %store output)
>>> +           (delete-paths %store (list output)))
>>> +     (and (equal?
>>> +            (map derivation-input-path
>>> +                 (derivation-prerequisites-to-build %store drv))
>>> +            (list (derivation-file-name input-drv)))
>>> +          (build-derivations %store (list drv))
>>> +          (delete-paths %store (list input-path))
>>> +          (not (valid-path? %store input-path))
>>> +          (null? (derivation-prerequisites-to-build %store drv)))))
>>> finding garbage collector roots...
>>> actual-value: #f
>>> actual-error:
>>> + (srfi-34
>>> +   #<condition &nix-protocol-error [message: "program 
>>> `/home/mcd/guix/nix/scripts/list-runtime-roots' failed with exit code 1" 
>>> status: 1] 380d4b0>)
>>> result: FAIL
>>
>> It’s seems to be the Guile running ‘list-runtime-roots’ that’s
>> segfaulting.
> 
> Actually it didn’t segfault: it exited normally, but with exit code 1.
> That shouldn’t happen.
> 
> Does “make check -j3” or so still trigger the problem?  If so, could you
> try collecting more details with:
> 
>   strace -f -o log -s 234 make check -j3
> 
> ?
> 
> TIA,
> Ludo’.
> 

In the meantime I updated the guix checkout and since my
.config/guix/latest points to the guix checkout, i checked 84bd92438 out
in another directory and ran
guix environment guix --ad-hoc strace
./bootstrap && ./configure --localstatedir=/var
./pre-inst-env strace -f -o log.fifo -s 234 make check -j3

make check -j3 doesn't always trigger it. But it did in the following 2
cases.

Your command creates a lot of output, so I piped straces output to
bzip2. It took very long so I aborted and retried but piped it onto the
host system to compress it. This is still takes very long when it
reached guix-daemon test. On top of it, I got a filesystem error which
resulted in a ro-remount, so I decided to interrupt the second try, too.
You can access the logs at
https://seafile.zfn.uni-bremen.de/d/7990941e630141309a58/


Martin