Ben Woodcroft <b.woodcroft@uq.edu.au> writes:

> On 17/08/16 09:45, Leo Famulari wrote:
>> On Wed, Aug 17, 2016 at 09:31:11AM +1000, Ben Woodcroft wrote:
>>>
>>> On 17/08/16 06:47, Leo Famulari wrote:
>>>> On Tue, Aug 16, 2016 at 11:45:16AM +0100, Marius Bakke wrote:
>>>>> I initially made this package on a foreign distro without "lapack" in
>>>>> inputs and have verified that dropping LAPACK makes the tests pass.
>>>>>
>>>>> I also found some other optional dependencies after digging around the
>>>>> source, as well as a recommendation to disable/enable asserts:
>>>>>
>>>>> http://dlib.net/dlib/config.h.html
>>>>>
>>>>> Ben, Leo: Can you try the following patch and see if that works for you?
>>>> Yes, this patch builds for me.
>>> Me too, although it appeared non-deterministic. I'm afraid I haven't time to
>>> see if this patch is suitable to push just now. Leo?
>> How did it appear non-deterministic to you?
> Just based on guix build --check:
>
> guix build: error: build failed: derivation 
> `/gnu/store/sxybcxw64q1ajzq6dysal75ffgq6238i-dlib-19.1.drv' may not be 
> deterministic: output 
> `/gnu/store/il57dcii4pzii11zlixjjxxxw699bg5x-dlib-19.1' differs
>
> I'm actually not sure, why does it say "may not be deterministic"? If it 
> builds twice and the second version is different, doesn't that mean it 
> is definitely not deterministic by counter-example, unless there has 
> been some leakage into the build container?
>
> I also tried building it with #parallel-build? #f and #parallel-tests? 
> #f. It worked in the first round but failed the second.
>
> Note also that the 'disable-asserts' phase should end with '#t'.

I've attached a patch with a #t in the disable-asserts phase, and also
deleting the 6MB static library.

Since `guix build --rounds=2` passes, is there any reason to delay this
patch? I've built this on GuixSD and a foreign distro and naively
verified that they are the same (the .so and all headers have the same
checksum; some of the recorded cmake input paths are different though).

Thanks,
Marius