Parallel builds in 3.1 branch?

[Mark Mitchell <mark at codesourcery dot com>]

Dave M. has asked me to consider applying my changes to support parallel
make to the GCC 3.1 branch so that we can speed up the build/test cycle.

I remember that there were some thoughts that my patch slowed things down
in serial.  Has anyone measured this?

The only change that seems a likely culprit is that I ordered the .java
files slightly differently:

all_java_source_files = \
    $(java_source_files) \
    $(built_java_source_files) \
    $(x_java_source_files)

The old loop did:

libgcj-@gcc_version@.jar: $(built_java_source_files) $(java_source_files) 
$(x_java_source_files)

There are only two built_java_source_files, though, so it's hard for
me to believe this really made things slower.

If nobody else has measured, I will do so.  Assuming that it really is
no slower than the old way, does anyone see a reason not to put this
on the GCC 3.1 branch?

--
Mark Mitchell                   mark@codesourcery.com
CodeSourcery, LLC               http://www.codesourcery.com