This is the mail archive of the
libc-alpha@sourceware.cygnus.com
mailing list for the glibc project.
Re: [Michael Deutschmann <michael@talamasca.wkpowerlink.com>] libc/1424: llseek symbol is attractive nuisance
- To: Michael Deutschmann <michael at talamasca dot wkpowerlink dot com>
- Subject: Re: [Michael Deutschmann <michael@talamasca.wkpowerlink.com>] libc/1424: llseek symbol is attractive nuisance
- From: Ralf Baechle <ralf at uni-koblenz dot de>
- Date: Sun, 28 Nov 1999 23:23:06 +0100
- Cc: Ulrich Drepper <drepper at cygnus dot com>, Andreas Schwab <schwab at suse dot de>, "GNU C. library Alpha Tester List" <libc-alpha at sourceware dot cygnus dot com>
- References: <m3puwycxv9.fsf@localhost.localnet> <%tebP44tBZ@khar-pern.talamasca>
On Thu, Nov 25, 1999 at 02:38:31PM -0800, Michael Deutschmann wrote:
> But an llseek() prototype *never* appears, even if _GNU_SOURCE is defined.
> And this llseek symbol means that autoconf's usual test will show the
> library as having "llseek" functionality, which it doesn't really.
>
> Remember that the missing prototype causes a calling-convention screwup,
> which makes llseek() dangerous. Think about what can happen when to an
> fsck program if the seek function cannot be trusted... it did, I'm told.
> T'so now maintains that you guys are ivory-tower lunatics over it.
The llseek() think is really painful. It was the cause of numerous
filesystems showing signs of aging too early ... I was the person to
report this llseek problem to Theo back in time. He wasn't pleased, to
say the least and probably quite a number of other people's filesystems
had already been ruined before I tracked this one down. I think a fix
and even if it's something as painfull as just removing llseek() completly
and thus breaking compatibility will all in all be less harmful.
Ralf