Re: [sdcc-devel] NetBSD on sparc64 regression failures - machine too slow?

[Erik P. <epe...@iv...>]

On Sat, 2 Jun 2012, Philipp Klaus Krause wrote:

> The regression tests are failing on NetBSD on sparc64. The exact failure
> varies, but it looks like this:
>
> libmullong                          (f:  0, t:  66, c:  2, b:  12366, t:
>   31052)
> literalop                           (f:  0, t: 291, c:  3, b:  17514, t:
>   17359)
> Warning: couldn't find crt0.rel
> ?ASlink-Warning-Couldn't find library 'r3ka.lib'
>
> ?ASlink-Warning-Undefined Global '_putchar' referenced by module ''
>
> ?ASlink-Warning-Undefined Global '_exit' referenced by module ''
>
> ?ASlink-Warning-Undefined Global '__modsint_rrf_s' referenced by module
> 'testfwk'
>
> ?ASlink-Warning-Undefined Global '__divsint_rrf_s' referenced by module
> 'testfwk'
> make[5]: ***
> [gen/ucr3ka/logic/logic_values_350_storage_none_type_char_attr_volatile.ihx]
> Error 1
> make[4]: *** [results/ucr3ka/logic.out] Error 2
> make[3]: *** [test-port] Error 2
> make[2]: *** [test-ucr3ka] Error 2
>
> I suspect that the machine is simply too slow to run the regression
> tests to completion and the next day's build pulls the library our from
> under the testing.
> Is there any other reason the library might disappear in the middle of
> testing? I'll see what happens when I do build sdcc and run the tests on
> the machine tomorrow (building sdcc took many hours of today, I expect
> the tests will complete sometime tomorrow), but I expect it to work fine.
>
> Philipp

A few days ago I asked Dave McGuire about low disk space or hitting a 
disk quota limit. He said that there were no quotas, the /homes volume had 
many gigabytes free, but /var was a little low (80 MB). He did some 
cleaning on /var in case it was running out of room in /var/tmp. But as 
you see that hasn't appeared to have changed anything.

I think your hypothesis sounds plausible. We could update the crontab 
entry so that the snapshot is only built every other day rather than 
daily. I also think it would be useful to add some timestamps to the 
regression tests.

   Erik