OSDev.org

I made a change to the article which removes some irrelevant information, do you agree with how it is now AJ?

My reason for this was, none of the environment variables set in the cross-compiler article are required for the installation of GMP or MPFR.

It may also be a possibility to recommend using the host OS's package management functionality, ports/packages for OpenBSD, pkgsrc for NetBSD, RPM/DEB's for Linux, etc.

They'll just need to ensure --with-gmp/mpfr are pointing to the correct location..

I'm torn, jnc100's method removes the requirement of building GMP/MPFR separately (and/or installing them system wide..).

Personally, it's not a big deal.. they're fairly small packages, in the end, the decision is up to you guys.

Yes - I just made a small change to the numbering now one of the points has been removed, but otherwise, it looks good.



Hmm...I suppose one disadvantage of this method is that GMP/MPFR have to be added to the GCC path each time you do this, whereas the first method makes them immediately available for when GCC 4.3.1+ is released (assuming that it doesn't require a later version of the other libraries, which IMHO is unlikely).

Perhaps an additional section could be added just before the links. Something like 'Alternative Method' (unless you feel this is promoting the first method and sidelining jnc100's way too much )

Cheers,
Adam

I completely agree. The additional disadvantage with my method is that we cannot guarantee that the gcc build system will always support such a method. The compile-each-package-separately method is a much more standard way of doing things.

Regards,
John.

Your solution almost work, but It seems that gcc is looking for the gmp headers at ../../gcc.4.3.3/gmp So I wrote



And it works! You have saved me the afternoon. Thanks a lot.
and... remember to install g++ before...

Wouldn't it have been easier to use --with-gmp and --with-mpfr rather than creating the symlinks?