[csw-maintainers] zeromq build failing on i386: Too many open files (signaler.cpp:300)

Romeo Theriault romeotheriault at opencsw.org
Thu Jun 21 01:11:04 CEST 2012


On Tue, Jun 19, 2012 at 9:06 PM, Dagobert Michelsen <dam at opencsw.org> wrote:

>>
>> This builds only libs 64-bit for other applications, client API etc.
>>
>> see http://sourceforge.net/apps/trac/gar/wiki/GAR%20Variable%20Reference
>> so you need
>>
>> BUILD64 = 1
>
> The variable reference lacked a description of
>  BUILD64_ONLY = 1
> which means to build only the 64 bit version. I guess this is what you need to
> verify the assumption on 64 bit.

Thanks Carsten and Dago, with the BUILD64_ONLY = 1 I was able to it to
build just on 64 and the same test still failed, so it's not a 32bit
vs 64bit dependent thing...

> You can see what flags are passed by using
>  mgar modenv

Nice, thanks for the info. This is good to know.

> Setting flags about optimization, ISA, memorymodel manually in CFLAGS is extremely
> dangerous as it interferes with the flags GAR sets for you. If you want to make
> changes please override/adjust the respective compile-specific variables:
>  http://sourceforge.net/apps/trac/gar/browser/csw/mgar/gar/v2/gar.conf.mk#L224
> Usually this means selecting the correct ISA and then adding things as necessary.

Ok.

>>> Also, assuming it does build as 64bit it seems that all applications
>>> that would need to link against it would need to be 64bit as well,
>>> right?
>
> Right.

Thanks for the verification.


> To go back to your initial problem: Does the increased filehandles make it to the process?

Bingo, it seems not! When I run the test manually (after up'ing the
filehandles) it returns without error, doesn't core and a returns a
return code of 0. Then when I bring the filehandles back down to 256
and run the test again it cores and errors out as it does during the
build run tests. Now to figure out why the ulimit settings in my
.bash_profile don't affect it when going through the build process.
Does the build process use 'sh' I wonder...

Will report back when I've found the culprit. Thanks for the very useful info.

Romeo


More information about the maintainers mailing list