On Mon, Jun 25, 2007, Mark Hatle wrote:
> In my environments, I setup the compiler and CFLAGS before calling
> configure w/ the right information to build 32, or 64 bit binaries.
> Having configure try to guess and pass in additional values is a royal
> pain. Passing in --libdir=/usr/lib64 (or equiv) is really the right
> answer. There shouldn't be any MARK64 or similar checks in configure or
> other components. If they care about type sizes, then they should use
> the size types, i.e. uint32 instead of "long", or they should check
> sizes using the autoconf variables and adjust things based on that.
>
> (One rare example. There is a mips64 target that comes back 32-bit in
> all of the tests. If the system checks for "64" in the arch name it
> WILL do the wrong thing. It needs to look at type sizes to determine
> what to do. Most software gets this right.)
I fully second your statements. As I said, from my POV the whole MARK64
is an ugly hack and really should be removed at all.
But as it is mainly a Linux thing and I (as a cross-platform Unix
hacker) do not care very much about the single Linux platform myself, it
would have been rude if _I_ just removed it. So, I just worked around
its problems by at least allowing us to disable it manually. But if we
can get rid of it at all -- even better!
So, anybody who is able to fully understand all Linux issues related to
this MARK64 stuff, please feel free to jump in and kick it out...
Ralf S. Engelschall
[EMAIL PROTECTED]
www.engelschall.com
______________________________________________________________________
RPM Package Manager http://rpm5.org
Developer Communication List rpm-devel@rpm5.org
