Trick or treat?

[Geert Bosch]

On Nov 1, 2004, at 00:47, Andrew Pinski wrote:
> There is one big bug which is already filed and is a regression from
> 3.4.2.  IA64-linux does not compile if you don't have libunwind
> installed already.

This regression was caused by H.J.'s patches. I went back and forth
with H.J. on this, trying different patches, installing new binutils
etc, without success, although I appreciate his effort to help.
The only workaround I've found so far is to manually set the
LD_LIBRARY_PATH to include the path with libunwind.

It seems that at least we will need a clear release note that explains:
    * why all these changes were needed for a bug fix release
      in the first place, and
    * why there are new binutils requirements (linker with --as-needed)
      and what the exact requirements are (just linker, or complete 
binutils)
    * what is the exact libunwind requirement
    * what are the installation requirements (hardcoded paths, env.vars)

Also, it is unclear (at least to me) wether user applications will be
affected or just the compiler? Can we still build a fully statically
linked binary and distribute it, or do we always depend on libunwind?
Similarly for dynamically linked apps/libraries.

Who understands the above issues well enough to produce a note
that is comprehensible by the average GCC user?

   -Geert (who still regards 3.4.3 as a "trick" on IA-64)