build failure on RHEL5

Thanks for reporting!

Afraid it still fails like this. Do you need me to try to fix it when I
have a chance?

procenv.c: In function ‘show_numa_memory’:
procenv.c:4295: warning: implicit declaration of function ‘numa_num_possible_n’
procenv.c:4296: warning: implicit declaration of function ‘numa_num_configured’
procenv.c:4298: warning: implicit declaration of function ‘numa_get_mems_allow’
procenv.c:4298: warning: assignment makes pointer from integer without a cast
procenv.c:4302: error: dereferencing pointer to incomplete type
procenv.c:4303: warning: implicit declaration of function ‘numa_bitmask_isbits’
procenv.c:4345: warning: implicit declaration of function ‘numa_free_nodemask’
procenv.c: In function ‘show_capabilities_linux’:
procenv.c:6738: error: ‘CAP_LAST_CAP’ undeclared (first use in this function)
procenv.c:6738: error: (Each undeclared identifier is reported only once
procenv.c:6738: error: for each function it appears in.)
procenv.c:6836: warning: implicit declaration of function ‘_show_capability’
procenv.c: In function ‘cap_get_bound’:
procenv.c:6897: error: ‘PR_CAPBSET_READ’ undeclared (first use in this function)

Gah - thanks for testing. Seems my test system didn't have the numa headers installed. Let me fix this properly... :-)

Hi Dave - please could you try the latest lp:procenv. This WFM using Centos 5.10 but I can't test the output is as expected on a NUMA system atm so if you could let me know, that would be much appreciated ;-)

James Hunt <email address hidden> writes:

> Hi Dave - please could you try the latest lp:procenv. This WFM using
> Centos 5.10 but I can't test the output is as expected on a NUMA system
> atm so if you could let me know, that would be much appreciated ;-)

I'm afraid it's still barfing on undefined CAP_LAST_CAP and
PR_CAPBSET_READ (on up-to-date RHEL 5.10, which should be effectively
identical). I guess the difference is that I have libcap-devel
installed. It's not immediately clear to me how to fix it.

If I build by pretending not to have sys/capability.h, the NUMA results
when bound to one core show a couple of unknowns, but I haven't checked
whether that's expected.

cpu:
  number: unknown of 4

...

memory:
  page size: 4096 bytes
  numa:
    api version: 1
    policy: MPOL_DEFAULT
    maximum nodes: unknown

Hi Dave - thanks for your patience. I believe I've now resolved the build issue if you could try lp:procenv again.

Could you provide any more details on the NUMA issue? procenv now uses the old numa 0.9.8 API for Centos 5 but I may be misusing that API. It would be great if libnuma actually had a little more documentation :-)

James Hunt <email address hidden> writes:

> Hi Dave - thanks for your patience. I believe I've now resolved the
> build issue if you could try lp:procenv again.

It builds OK now, thanks.

> Could you provide any more details on the NUMA issue? procenv now uses
> the old numa 0.9.8 API for Centos 5 but I may be misusing that API. It
> would be great if libnuma actually had a little more documentation :-)

It looks as if it's just a result of ! HAVE_SCHED_GETCPU and
LIBNUMA_API_VERSION. I should have looked before.

Right - seems that glibc on rhel5 is too old to have sched_getcpu(). I'll put in a crutch to work around that...

I've just updated lp:procenv - please could you check that it now shows the correct CPU details for you. Thanks!