Re: [Jfs-discussion] Broken configure script for RHEL4?

[Dave K. <sh...@li...>]

On Thu, 2008-07-17 at 10:41 -0400, Dan Swartzendruber wrote:
> Sorry if this is the wrong place for this.  I was trying to install JFS on
> a clarkconnect server, which is based on RHEL4.  I got a source RPM and
> did a build, and it seemed to go okay.  However, I was unable to run
> jfs_mkfs, which would complain about this:
> 
> The specified disk did not finish formatting.
> 
> After much experimenting, I discovered that I couldn't create a filesystem
> larger than 2GB or so.  After many printfs added here and there, I
> discovered it was an error from fseeko due to EINVAL (negative offset). 
> But, the disk_offset parameter to the routine in question is a 64-bit
> integer.  I looked at the configure script and tried a hunch.  Here is the
> output from the configure script:
> 
> checking for _FILE_OFFSET_BITS value needed for large files... 64
> checking for _LARGEFILE_SOURCE value needed for large files... no
> 
> I changed _LARGEFILE_SOURCE to 1 in config.h, rebuilt and tried again, and
> it works.  I don't know enough about this to know why this is happening
> (apparently, the script is wrong on RHEL4?)

The problem was introduced when I build the source tarball using
autoconf-2.61.   Apparently, _LARGEFILE_SOURCE isn't really needed with
the header files shipped with newer versions of glibc, but it breaks
when built with glibc-2.3.  I have a fix in cvs to work around this by
defining _LARGEFILE_SOURCE in libfs/devices.c, but I've been negligent
and haven't published jfsutils-1.1.13 yet.

Thanks for reporting the problem, and I'm sorry that I put you through
so much trouble to track down the problem which I should have fixed with
a new release.

Shaggy
-- 
David Kleikamp
IBM Linux Technology Center