[OpenAFS-devel] butc crash on linux/amd64 + xstat/afsmonitor

[Rainer Toebbicke]

Jeffrey Altman wrote:
> Note that there were similar problems in
> 
>   src/kauth/rebuild.c
>   src/lwp/test/selsubs.c
> 
> 

... and src/xstat/xstat_cm_test.c,
openafs/src/xstat/xstat_fs_test.c,
src/afsmonitor/afsmon-output.c,
afsmonitor/afsmon-parselog.c

For those I've got a fix ready. Note that xstat/afsmonitor return 
hopeless nonsense when they don't refuse to collaborate at all due to 
the fact that 'struct timeval' is now 2 times 8 bytes. Actually the 
only "honest" player is the primitive xstat, afsmonitor seems to be 
more than obsolete.

I've fixed that temporarily by coercing everything to afs_int32 like 
the RPC definition actually states, as otherwise you'll only be able 
to run xstat among machines not only of the same release but also of 
sufficiently close architectures.

afs_int32 is admittedly a little brutal given that the absolute 
timestamps are going to pose problems as of the early morning of 19 
Jan 2038 (in Geneva) that way. One could gain some thinking time by 
defining tv_sec unsigned if we're not all on quantum machines by then.

A proper way would of course be to cast everything consequently to 
64-bit, but why 64 and not 128 next year, where is the RPC support 
(out of curiosity: do mixed big/little endians exists among 
processors/compilers?) and what about the backwards compatibility to 
the old 32-bitters. For me the practical argument prevails dispite the 
obvious caveats.

 > Any place where there is casting to (time_t *) is an indication that
 > there is probably a problem lurking behind.

We possibily met a few years ago on one of the Transarc events (if 
not, sorry!) so I believe I can safely judge that we both won't be 
*directly* affected :-) My cynical me points out that the now 
youngsters may welcome a possible gain in expressed interest for them 
in the industry near or after the end of their career approaching that 
date. Oh well...

oops - chatting away off-topic!

-- 
=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=
Rainer Toebbicke
European Laboratory for Particle Physics(CERN) - Geneva, Switzerland
Phone: +41 22 767 8985       Fax: +41 22 767 7155