Bug#790571: please give-back capnproto on mipsel

Control: tags +confirmed

Dejan, Arturo, thanks for looking into this. Sorry I've been so slow
to get back to you.

Arturo, I understand you've followed up with upstream regarding the
affected tests & Kenton's helping you out with some experimental
patches to disable the failing tests. I'd be interested to know how
that works out, but I do worry that by disabling the tests we'll be
potentially glossing over a real issue.

Dejan, I think you might be onto something with that 4k buffer but let
me look into it a little. Are you aware of any porterboxes with
similar setups to mips-aql-02 + mipsel-manda-0{1,2}?
https://db.debian.org/machines.cgi?host=eder (Loongson 2E) looks like
it might be a promising candidate from a quick google around.

I'm guessing this is likely what's causing builds to fail on several
other archs too.

Cheers,
Tom

On Fri, Jul 17, 2015 at 8:20 AM, Dejan Latinovic
<[email protected]> wrote:
>
> Hi,
> I have tested capnproto on a few local machines.
> Initially, build failed on all boards.
> On different MIPS boards, different tests were failing.
>
>
> AsyncUnixTest.WriteObserver fails if the kernel PAGESIZE is larger that 4k.
>
> After I reduced PAGESIZE to 4096 on CI20, all tests passed.
>
> The solution could be to increase buffer size:
>> char buffer[4096]
> (src/kj/async-unix-test.c++ +416)
> I had increased it to 16384 and tried it on Loongson 3A
> (PAGESIZE is 16k, same board as mipsel-manda-01, mipsel-manda-02),
> all test passed.
> We should keep on mind that pagesize on some MIPS board is up to 64k.
> This solution should be discussed upstream.
>
>
> On EdgeRouter Pro (mips-aql-02),
> these two test failed:
>> [  FAILED  ] 2 tests, listed below:
>> [  FAILED  ] AsyncUnixTest.SignalWithValue
>> [  FAILED  ] AsyncUnixTest.SignalWithPointerValue
>
> I will do further investigating.
>
> Best Regards,
> Dejan



-- 
Tom Lee / http://tomlee.co / @tglee