Bug#995867: xtensor: mipsel FTBFS (relocation truncated to fit: R_MIPS_CALL16)

[Drew Parsons]

On 2021-10-27 14:03, Adrian Bunk wrote:
> On Wed, Oct 27, 2021 at 11:41:39AM +0200, Drew Parsons wrote:
>> ...
>> Since these are runtime tests (compiled fresh for the test), I like 
>> the idea
>> of keeping -march=native, since it will emulate what users will 
>> experience
>> when they use xtensor in their own local projects.
>> ...
> 
> The flakiness in build results we are seeing on mips* due to that
> can be a problem in general. You just got first-hand experience on the
> impossibility of debugging a problem that happens only on some buildds.
> 
> ARM builds are currently a variety of hardware, and there are literally
> dozens of cpu feature extensions some whom are implemented on a given 
> cpu.

I'll have keep in eye on that.  fenics-dolfinx for instance is failing 
on armhf, though possibly not because of -march=native.

> BTW: xtensor-blas/mipsel did just FTBFS due to this problem:
> https://buildd.debian.org/status/fetch.php?pkg=xtensor-blas&arch=mipsel&ver=0.19.1-6&stamp=1635332411&raw=0


Good catch.  I'll patch it up also.

Drew