showstopper in 05.12 snapshot

[Thomas Mike Michlmayr]

On Sat, Dec 11, 1999 at 19:34:17 +0100,
Thomas Mike Michlmayr <mike at cosy.sbg.ac.at> wrote:
[...]
> does anyone have any idea of what's going on here? config files can be
> provided if necessary.

i was able to circumvent the problem. turns out that "--with-largefiles",
in comibnation with "--enable-tagged-hash" on solaris 2.6 was the culprit.

barry bouwsma, lying on the floor next to me, remembered enough of his
dbz hackery long long ago, and suggested looking into dbz.c. and the guess
is that the _longlong_t that solaris defines OFFSET_T to seems to cause
some subtle problems.

compiling without largefiles support fixed the problems with dbz, but
showed another problem.

the newly compiled innd crashed on startup. i tracked the crash into
overview (buffindexed.c: in buffindexed_groupadd, the GROUPfind ended
innd). buffindexed overview seems to have some issues when swtiching 
between 32 and 64 bit file support, as it uses OFFSET_T quite heavily. 
i was able to get innd running again by changing all OFFSET_Ts to 
_longlong_t.

IMHO it's not a good idea to have the fileformat depend on the largefiles
(can we use 64 bit counters all the time?), as i suspect a large number
of inn-users starting out with 32 bit filesystemcalls, and later 
recompiling with 64bit fs-calls.

it looks like CNFS and timecaf already handle this gracefully (at least
i'm using them and those files had no problems).

-- 
Thomas 'Mike' Michlmayr can not assert the truth of all statements in 
this article and still be consistent. <mike at cosy.sbg.ac.at>



-- Attached file included as plaintext by Listar --

-----BEGIN PGP SIGNATURE-----
Version: 2.6.3i

iQCVAgUBOFYjCQGXiSxHZWmZAQEhIwP/XXJmnAORoyHLJDg1O+R6ihaiRIl2XzNV
1GUWDQLyiIvxMU4wDaDRzXGugR3GNIqD4lN2n8SFNinFS7U+6rUlYNNNgNg0BWH5
eLd318Pp+iif7z1A5Zs7KtGRpBlfhSx0xoOaCy+TBCJX9xZgMAsKqk/uAJUqDRbF
I6X9ZdLWxDw=
=vE4A
-----END PGP SIGNATURE-----