This is the mail archive of the
libc-alpha@sourceware.org
mailing list for the glibc project.
Re: [PATCH] Bump up stdio bug22 timeout from 30 seconds to 60 seconds
On Thu, May 17, 2012 at 4:05 PM, David Miller <davem@davemloft.net> wrote:
> From: "Carlos O'Donell" <carlos@systemhalted.org>
> Date: Thu, 17 May 2012 13:51:56 -0400
>
>> On Thu, May 17, 2012 at 1:48 PM, Chris Metcalf <cmetcalf@tilera.com> wrote:
>>> Thanks, I didn't know about the "-ENV" mechanism. ?I suspect it's
>>> reasonable to leave the global commit I did as-is, but let me know if you
>>> think I should revert it in favor of the tile-specific hack.
>>
>> The global change is fine with me. IMO as long as a machine doesn't
>> require some ridiculously long timeout I'm happy bumping the default
>> test timeout rather than adding code per-machine to tweak the timeout.
>
> While we're on this topic, any objections to the following?
>
> This happens on sparc 32-bit because the 32-bit code generation for
> all of the 64-bit long long operations in sha512 is terrible. ?I don't
> anticipate this being fixed in the near future, and it's silly for the
> testsuite to have one failure because of this.
>
> I have some plans on how to solve this in other ways. ?For example, I
> might just make a sparc assembler implementation of sha512 for 32-bit
> sparc. ?When targetting sparcv9 we can really improve things
> substantially since we have access to 64-bit integer registers and
> operations in certain registers even when 32-bit.
>
> 2012-05-17 ?David S. Miller ?<davem@davemloft.net>
>
> ? ? ? ?* crypt/sha512c-test.c (TIMEOUT): Increase to 32.
>
> diff --git a/crypt/sha512c-test.c b/crypt/sha512c-test.c
> index c829242..060e935 100644
> --- a/crypt/sha512c-test.c
> +++ b/crypt/sha512c-test.c
> @@ -58,6 +58,6 @@ do_test (void)
> ? return result;
> ?}
>
> -#define TIMEOUT 6
> +#define TIMEOUT 32
> ?#define TEST_FUNCTION do_test ()
> ?#include "../test-skeleton.c"
I'm fine with this. Generally 30-60 seconds is reasonable for a small
test timeout, at least given the number of glibc tests.
Anyone who says the purpose of these timeouts is to catch performance
regressions is dreaming in technicolor.
Cheers,
Carlos.