gcj 3.0 post-mortem

Jeff Sturm jsturm@one-point.com
Sat Jul 7 14:39:00 GMT 2001


On 4 Jul 2001, Tom Tromey wrote:
> - Ideally we would get the Solaris port working.  It is unclear if
>   Andrew has the time to do that though.

I spent some time on it just prior to the release.  The signal handling
can be solved either by requiring sjlj exceptions, or defining
MD_FALLBACK_FRAME_STATE_FOR on solaris.

Unfortunately, the former is undesirable, and the latter difficult.
Prerequisites to implementing MD_FALLBACK_FRAME_STATE_FOR are:

- we can reliably detect when executing in a signal handler, and
- the sigcontext can somehow be obtained from the call stack.

This is without imposing any special requirements on the signal handler.

The details on just how to do this vary by hardware and OS, and
probably different releases of the OS as well.  It will require some
reverse engineering on platforms where kernel/libc source is not
available... I don't believe the stack layout of a signal handler is
specified by the platform ABI.

Jeff



More information about the Java mailing list