pocl-devel

[pocl-devel] No fdatasync on macOS

[Andreas K. <li...@in...>]

Hi all,

Here's the next macOS build failure on macOS on POCL 1.1:

https://travis-ci.org/conda-forge/pocl-feedstock/builds/351854461

fdatasync() doesn't seem to exist.

On a related note, I recall one of you commenting that you might be
interested in a macOS build slave. I might be able to help with
that. It's a Sandy Bridge Mac Mini with 8 GiB RAM that's sitting in a
lab here. It's not hugely powerful, but it should be able get the job
done.

Andreas

Re: [pocl-devel] No fdatasync on macOS

[Andreas K. <li...@in...>]

Andreas Kloeckner <li...@in...> writes:
> Here's the next macOS build failure on macOS on POCL 1.1:
>
> https://travis-ci.org/conda-forge/pocl-feedstock/builds/351854461
>
> fdatasync() doesn't seem to exist.
>
> On a related note, I recall one of you commenting that you might be
> interested in a macOS build slave. I might be able to help with
> that. It's a Sandy Bridge Mac Mini with 8 GiB RAM that's sitting in a
> lab here. It's not hugely powerful, but it should be able get the job
> done.

On yet another related note, you may want to upstream this patch:

https://github.com/conda-forge/pocl-feedstock/pull/15/commits/20e2a4a3137e57470b965e2ab9be9cd4637cf020

(Credit goes to Isuru Fernando, who figured out what the problem was and
created the patch.)

Andreas

Re: [pocl-devel] No fdatasync on macOS

[Michal B. <Fra...@ru...>]

Hi,

> fdatasync() doesn't seem to exist.

.. and yet, CMake finds it:

-- Performing Test HAVE_FDATASYNC
-- Performing Test HAVE_FDATASYNC - Success

so either the CMake test gets it wrong, or it exists and there is some
missing include somewhere. Plus there are more problems:

/Users/travis/miniconda3/conda-bld/pocl_1520730988784/work/pocl-1.1/lib/CL/pocl_timing.c:114:10:
note: forward declaration of 'struct tm'

/Users/travis/miniconda3/conda-bld/pocl_1520730988784/work/pocl-1.1/lib/CL/pocl_timing.c:136:3:
warning: implicit declaration of function 'gmtime_r' is invalid in C99
[-Wimplicit-function-declaration]

> On a related note, I recall one of you commenting that you might be
> interested in a macOS build slave. I might be able to help with

There's an open PR for it by Jeff Hammond, with a Travis build bot
setup, so i think your Mac Mini will *probably* not be required. What
we need more than buildbots is maintainers (or at least someone who
occasionally makes it work on Mac).

Cheers,
-- mb