Cross-compilers co-residence question

[Geoff Keating]

Bill Webster <billw@lineo.com> writes:

> I have been working with two cross-compilers, an Intel version and a
> PowerPC version. I have installed both of them into /usr/local. While do
> some evaluation of how the packages co-reside I discovered that there
> are 5 binaries which are unique exe's but which exist in the same
> directory.
> 
> When I build the Intel cross-compiler and install it the five files are,
> and they all appear to be Intel specific files:
> /usr/local/bin/cpp - executable

This is actually a bug.  Both of the cpp copies should be named like
powerpc-eabi-cpp.  This is fixed in the current CVS.

> /usr/local/bin/gcov - executable

This doesn't differ between targets (I think).  At least, it doesn't
differ much :-).

> /usr/local/lib/libiberty.a - library

This shouldn't differ at all.  It's only for the host.

> /usr/local/lib/libbfd.a - library
> /usr/local/lib/libopcodes.a - library

These are more interesting.  If you knew in advance you were going to
do two cross-compilers, you'd build a bfd with support for both
platforms.  However, it doesn't matter; these libraries are for the
host, and in practise are not needed once the binutils and gdb are
built.

> When I build the PowerPC cross-compiler and install it the above five
> files are replaced by PowerPC versions. Now if I am correct the first
> two are created from the build of gcc but the remaining 3 library files
> come from glibc.

No, all from gcc.

> Is this apparent overlap of binaries a problem?
> Is it safe to either version of these files regardless of the desired
> target?
> Are these files in the correct location? or do they need to be placed
> into a platform specific directory?

These are all fine (except for the cpp issue).
-- 
- Geoffrey Keating <geoffk@cygnus.com>