Peter Mount writes:
> Thats correct. Basically you are talking of something like PL/Java. The
> Java side would be simple, but its linking the JVM to the backend that's
> the problem.
I've tried that recently, here's how it looks as far as Linux JVMs go:
* Kaffe has a very polluted name space. Calls to its own functions get
resolved to PostgreSQL, and vice versa. Crash and burn result. The Kaffe
folks have admitted that this should be fixed but I didn't look farther
yet.
* The Sun/Blackdown JVM didn't work at all (not even 'java -version')
until I upgraded my libc. Then a simple test run crashes with an "error
external to JVM"; at first it looked like a segfault when referencing a
string constant. In gdb I saw myself faced with about 10 threads running
when nothing was going on yet, at which point I was too exhausted to
proceed.
* IBM's offering didn't work at all. I don't recall the problem anymore
but I think it didn't even link correctly.
So currently I don't see how this could become a mainstream project, let
alone across platforms.
> I know some people think this would slow the backend down, but it's only
> the instanciation of the JVM thats slow, hence the other reason fork() is
> holding this back. Ideally you would want the JVM to be running with
> PostMaster, and then each backend can then use the JVM as and when necessary.
But how do the other languages cope? Starting up a new Perl for each
backend can't be so cheap either.
--
Peter Eisentraut peter_e@gmx.net http://yi.org/peter-e/
