code review 7040048: exp/gotype: skip tests on freebsd/arm

exp/gotype: skip tests on freebsd/arm

The freebsd/arm builder has such poor IO bandwidth that it cannot complete this test within the 120 seconds allocated. Temporarily disable the test until a better kernel is available.

[dave_cheney.net, 2013-01-03 12:56:22]

Hello minux.ma@gmail.com, gri@golang.org (cc: golang-dev@googlegroups.com),

I'd like you to review this change to
https://go.googlecode.com/hg/

[dave_cheney.net, 2013-01-03 13:00:17]

Robert, Minux,

I would like to skip the test exp/types on freebsd/arm so we have a working
builder.  I'm actively tracking the progress of the FreeBSD-CURRENT kernel and
hope to remove this exception as the performance of the arm kernel improves.

Many Thanks

Dave

[minux1, 2013-01-03 13:45:28]

On Thu, Jan 3, 2013 at 9:00 PM, <dave@cheney.net> wrote:

> I would like to skip the test exp/types on freebsd/arm so we have a
> working builder.  I'm actively tracking the progress of the
> FreeBSD-CURRENT kernel and hope to remove this exception as the
> performance of the arm kernel improves.
>
Sure. LGTM.
At least go/types is now passing with our extensive debugging prints.

[dave_cheney.net, 2013-01-03 22:04:04]

> Sure. LGTM.
> At least go/types is now passing with our extensive debugging prints.

Thank you. Here is a comparison to another raspberry pi running this
test on linux/arm (from a few weeks ago)

ok  	exp/types	4.606s

and a comparison with an older arm5 (the freebsd/arm image does not
support hardware floating point) host from a few days ago.

ok  	exp/gotype	22.849s

Yet on this host, the test is a minimum of 100 seconds, probably due
to it's very slow IO, and assertion enabled kernel.

Does anyone else have any comments on this proposal ?

Cheers

Dave

[dave_cheney.net, 2013-01-04 11:24:08]

I was able to jerry rig an external USB hard drive to this host, which has much
better throughput. The time to run the test in exp/gotype was ~55 seconds.
Hopefully this will mean the builder passes the next build that is committed.

[rsc, 2013-01-04 15:11:02]

A better solution would be to reduce the test cost for all systems, by
excluding some when testing.Short() is true.

Russ

[dave_cheney.net, 2013-01-04 22:57:55]

It looks like the overnight builds passed. I attached a USB hard drive to the
host and this improved Io throughput enough to pass the test. 

Ill abandon this proposal. 

On 05/01/2013, at 2:11, Russ Cox <rsc@golang.org> wrote:

> A better solution would be to reduce the test cost for all systems, by
> excluding some when testing.Short() is true.
> 
> Russ

[dave_cheney.net, 2013-01-05 03:12:15]

*** Abandoned ***