[collectd] AIX port :D

[Florian Forster]

Hi Andrés,

On Fri, Nov 20, 2009 at 05:37:06PM +0100, Andrés J. Díaz wrote:
> We are pleased to announce the collectd port for AIX (was tested on
> AIX 5.3 and AIX 6).

wow, great :) I have pulled in the patches and put them in an extra
branch called “ms/aix”.

> The patchset was  forked from collectd branch 4.8.1, so we do not
> known if it's work fine for older versions.

I've rebased the patches to the master branch (i.e. the development
branch) without any conflicts.

> aix-configure.patch:
>       Add -lm in NaN 0.0/0.0 test in configure, because isnan is a
> library function in AIX. It could be another test...

I'll take a look at this. Generally I try to avoid checks of the form
“if operating system is AIX, then …”.

> aix-version-gen.patch:
>       Fix echo (echo binary do not support -n switch)

Ouch, I though that was supported everywhere..

> aix-src-Makefile.patch:
>       Flags for linking:  -Wl,-bexpall,-brtllib
>       Link with -lm because isnan is in libm

Adding “-lm” would ideally be handled by the NAN-check in the configure
script, too. What do those other linker arguments do? It sounds like
something that should go into the README file.

> Hacks (a feedback about how to improve these hacks will be welcome :D):
> 	axi-src-common.patch:
> 		AIX has ntohll and htonll defined, It could be a config check...

Yeah, a configure check is probably the best way here. I'll look into
that one, too.

> 	aix-scanner-parser.patch:
>  		AIX has TRUE and FALSE defined in sys/types.h (sic)

I don't see a problem with that solution, so in my opinion it can stay
like this (i.e. using “BTRUE” and “BFALSE” to avoid the other
definitions).

> 	aix-strange-stat.patch:
> 		The stat function when the file do not exists return
> 		EINVAL, instead of  ENOENT, but it's only fails on
> 		collectd code, a single test code return the good
> 		value... We do not known how... :?

Hm, weird. Did you try setting “errno” to zero before calling stat(2)?

> 	aix-notested-multimeter.patch:
>                 Fixed multimeter plugin adding a dirty and ugly cast,
>                 we cannot check this patch properly...
> 		-ioctl(fd, TIOCMBIC, &rts);
> 		+ioctl(fd, (int)TIOCMBIC, &rts); /* XXX */

Sounds like that “TIOCMBIC” is somehow weird. What is the error message
you see without the cast?

> 	aix-notested-memcached.patch:
> 		AIX don't have MSG_DONTWAIT. The memched module could be
> 		works, but we cannot tested yet.

That looks okay for me, though I'll probably change that to
“#ifndef MSG_DONTWAIT” instead. As I said, I try to avoid OS-based
decisions.

> The hard work for these patches was done by my colleage Manuel
> Sanmartin, so credits for him please :)

I have that name as author for all the patches. If I messed up, please
let me know so I can fix the commits. And please tell Manuel a big thank
you from me, I really appreciate his work even though I haven't used AIX
myself yet. (People who have tend to say that I'm lucky ;)

Regards,
—octo
-- 
Florian octo Forster
Hacker in training
GnuPG: 0x91523C3D
http://verplant.org/
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 189 bytes
Desc: Digital signature
Url : http://mailman.verplant.org/pipermail/collectd/attachments/20091123/5e09f0ac/attachment.pgp