Marius Bakke <mbakke@fastmail.com> writes:

> Ludovic Courtès <ludo@gnu.org> writes:
>
>> Hi!
>>
>> Marius Bakke <mbakke@fastmail.com> skribis:
>>
>>> * gnu/packages/ninja.scm (ninja)[arguments]: Set #:tests? #f on armhf.  Adjust
>>> CHECK-PHASE accordingly.
>>> ---
>>>  gnu/packages/ninja.scm | 20 +++++++++++++++-----
>>>  1 file changed, 15 insertions(+), 5 deletions(-)
>>>
>>> diff --git a/gnu/packages/ninja.scm b/gnu/packages/ninja.scm
>>> index adb236484..cff2ac38b 100644
>>> --- a/gnu/packages/ninja.scm
>>> +++ b/gnu/packages/ninja.scm
>>> @@ -42,7 +42,15 @@
>>>      (build-system gnu-build-system)
>>>      (native-inputs `(("python" ,python-2)))
>>>      (arguments
>>> -     '(#:phases
>>> +     ;; FIXME: On armhf-linux, Subprocess.TestWithLots fails like this:
>>> +     ;;   *** Failure in src/subprocess_test.cc:241
>>> +     ;;   ExitSuccess == procs[i]->Finish()
>>> +     ;; See <https://hydra.gnu.org/build/2515522/nixlog/1/raw>.
>>> +     ;; Disable for now.
>>
>> Sounds reasonable.  Could you also report it upstream?  :-)
>
> Actually, now that I'm more awake, I found the problem.  On most
> machines this test is skipped because it requires 1025 open files.
> However, on Hydras armhf machines, this limit is apparently higher.
>
> The problem is that the test calls out to /bin/echo.  I was able to
> reproduce it on x86_64 simply by raising the softlimit to 4096.
>
> So, I pushed a one-line fix as d0de4c40e24ee52e4382ee2ac82088932ab49099.

Derp.  Apparently I wasn't awake enough and managed to patch the wrong
file when making the commit.  The followup in
6d796220c417a34100c282a7fa1f9afb6ac00339 fixes it, sorry for the noise.