Re: Building on MacOS X

[Brian Avery <avery.brian@gmail.com>]

[-- Attachment #1: Type: text/plain, Size: 3664 bytes --]

Hi,

A couple of comments even though I'm coming late to the discussion.

So, from what I've understood from the above, the main issue with the
docker approach to building a yocto/oe image on the mac is that it was much
slower...  Here's  some numbers that don’t quite agree with that assertion:

I ran a couple of tests, some on my linux box and some on my mac laptop.
The tests involved building core-image-minimal for a qemux86 target. All
the downloads were in place. There was no sstate, parse cache, nor was
there an existing tmp directory before the tests were run.

My linux box has a Intel(R) Xeon(R) CPU E5-2699 v3 @ 2.30GHz , 36 cores.
My mac laptop has a Intel(R) Core(TM) i7-4870HQ CPU @ 2.50GHz.

The test is "$ time bitbake core-image-minimal"

Here are the results of the tests:

Linux box Ubuntu 14.04

clock : 30m24.024

user  : 229m12.324

sys    : 20m10.188

—-

Linux box Ubuntu 14.04 running inside docker 1.12.3 using crops/poky:latest

clock : 30m37.66

user  : 231m35.984

sys    : 31m20.204


——

Linux box Ubuntu 14.04 running inside docker 1.12.3 using crops/poky:latest

I set the following in my local.conf file:

BB_NUMBER_THREADS=“2”

BB_NUMBER_PARSE_THREADS=“2”

PARALLEL_MAKE=“-j 2”

AND I constrained docker to 2 cpu cores and 8 gb of ram (This is what I
have my mac laptop set to).

clock : 127m8.523

user  : 201m32.468

sys    : 19m54.052


—

Mac OSX laptop - running docker 1.12.5 2 cpus, 8gb ram

I set the following in my local.conf file:

BB_NUMBER_THREADS=“2”

BB_NUMBER_PARSE_THREADS=“2”

PARALLEL_MAKE=“-j 2”


clock : 99m31.190

user  : 137m40.400

sys    : 18m38.650


——


So, from the above, it looks like my mac is actually faster when it has the
same number of cores and memory.  I haven’t seen any particular slowdown
with running docker linux programs on the mac other than those caused by
the difference in horsepower between my build server and my laptop.


Thanks,

Brian Avery

an Intel employee


p.s. Andrea, would you mind replying with how you changed the docker run to
make loopback work? I’d like to add it to the docs. Also, if you could
point me at what layers you used to make resin, I’d like to give that a try
as well.  Right now, we are providing a bare bones environment but I’d be
happy to write up a howto for inheriting from our images to customize your
own for special purposes (like doing builds that require loopback mounts,
for instance).







On Mon, Jan 16, 2017 at 3:19 AM, Burton, Ross <ross.burton@intel.com> wrote:

>
> On 14 January 2017 at 19:45, Roger Smith <roger@sentientblue.com> wrote:
>
>> Is Building Yocto project on a POSIX system, a desire for the Yocto
>> project? It would allow support on all bsd UNIX’s including macOS
>>
>
> Making OE itself work isn't rocket science - fix a few Linuxisms in
> bitbake, port pseudo to macOS.
>
> The hard bit is then convincing the hundred-odd recipes that are often
> Linux-centric if not Linux specific to build under something that isn't
> Linux.  My ross/darwin branch (from before the security changes) has a
> patch to gmp as 'echo' has different semantics. unlink() has different
> error codes between macOS and Linux.  There's a very long tail of
> differences that will need patching and testing.
>
> But if this is something you care about, patches welcome!
>
> Ross
>
>
>
> --
> _______________________________________________
> yocto mailing list
> yocto@yoctoproject.org
> https://lists.yoctoproject.org/listinfo/yocto
>
>

[-- Attachment #2: Type: text/html, Size: 7796 bytes --]