Nightly Japi comparisons

[Tom Tromey <tromey at redhat dot com>]

I recently set up nightly gcj builds here.  I didn't use some of our
internal infrastructure (lazy, etc), so the regression output is not
yet automated.  But the libgcj / JDK API comparison is.  Stuart
Ballard hosts the results on his japi page:

    http://japi.sab39.org/

I think the results here will generally be less than 24 hours old, at
least for libgcj (unless I go on vacation or something -- hopefully
I'll move the infrastructure before that).

Note that the classpath results on that page are currently a bit
stale.  That should be fixed soon, like this week.


The libgcj results report the cvs trunk, and include GNU Crypto.
We're just shy of 50%:

            good   minor   bad   missing  abstract addition
    Total:  49.7%  2.72%  0.36%   47.2%       0.01%


Thanks to Stuart for writing japi and hosting this page.  This is a
nice tool, since it lets us see our API divergences very easily.

Tom