FW: Java vs. Dwarf2 EH Unwinder

[Bryce McKinlay]

>
>
>  Andrew> Surely you'd need to pass the entire processor state to the
>  Andrew> _Unwind_RaiseException() variant.
>
>  Rich> Indeed.  I would be willing to create a variant that took a
>  Rich> sigcontext* or something.
>

That would be really good. Perhaps the unwinder could expose its 
_Unwind_Context struct somehow and libjava could copy the registers into it.

>  Andrew> This would only work if the kernel saved the entire processor
>  Andrew> state in the sigcontext and then called the signal handler.  This
>  Andrew> does happen on some architectures, but not all, and therefore
>  Andrew> it's not a totally general solution.
>
>  Rich> There has to be enough information for the kernel to restore
>  Rich> enough state.  What do you have in mind as an example of
>  Rich> not-enough-state being saved?
>
>For example, ia64 linux only saves the scratch registers on signal
>delivery.  Thus, some preserved registers may live on the signal stack
>and you can't just short-circuit out of the signal handler.
>

OK, but it sounds like ia64 is something of a special case for 
unwinding, and it doesn't have any problem with unwinding signal frames? 
The motivation here is to get other platforms which use the dwarf2 
unwinder working for Java - the ones that are working already can 
continue to use their existing mechanism if it makes sense.

regards

Bryce.