gcj 3.0 post-mortem

Wed Jul 4 11:01:00 GMT 2001

Benjamin Kosnik had the good idea that we should look at how the gcj
3.0 release compared to our expectations.
Overall I think we did ok.
We had planned to have it working on Linux (x86, IA-64, alpha, and
PPC), plus Solaris. I believe in the end the IA-64 Linux and Solaris
ports did not work. This happened because the new EH code required
more work than we had anticipated, and our lone Java EH expert, Andrew
Haley, was scheduled to work on other projects during the window we
needed his expertise.
We had planned to get the new Calendar code into 3.0, but I never
found the time to do it.
We had planned to have in-tree builds working, but I made a mistake
and failed to test `make install'. This is fixed for 3.0.1 already.
My recommendations:
- I'm not really aware enough of the entirety of the EH saga to really
 make recommendations there. I'm guessing there was a (probably
 minor) communications gap between the gcc developers and the Java
 developers about Java's requirements for the new EH code. Andrew,
 can you say more here? What can we do differently next time to
 avoid a suprise like this?
- For the Calendar code, I should have asked for a volunteer to do the
 merge. In fact I think there was a volunteer and I told him that I
 would be doing it. My recommendation is that I say "yes" more
 frequently to volunteers.
- The in-tree build problem was also mine :-(. That one I classify as
 a simple mistake; I don't think there is a procedural issue to
 address.
I did think that there were bumps in the process, mostly relating to
the "non-critical" nature of the Java code. Many of these bumps have
already been ironed out, and I think this will happen without any
special effort as the Java front end gains in acceptance in gcc-land.
For 3.0.1:
- Ideally we would get the Solaris port working. It is unclear if
 Andrew has the time to do that though.
Anybody else have comments? Is there something I missed? I'm unaware
of any other to-do items for 3.0.1. Are there more?
Tom