Re: Exception handling and embedded `gcj' (fwd)

[Godmar Back <gback at cs dot utah dot edu>]

oops, first attempt went to old mailing list alias...
I'm resending this mail:

 Hi Jon,

since your email from February this year (the one to which I'm replying)
was brought up in the recent discussion regarding setjmp/longjmp based exception
handling, I reread your mail from February in which you report about how you
implemented a hybrid scheme to deal with exceptions in a way that's better suitable
for embedded environment.

Now that I have some experience with dwarf and related issues, I'm wondering whether 
the overhead you're paying is indeed only three register saves as you enter a try block.
You described that your exception handler assumes that all registers are clobbered 
upon entry; does this not imply that you will have to spill them at any point in 
the cfg where an exception could occur?  If so, shouldn't that also be counted
as run-time overhead?  Also, do you deal with null pointer exceptions in hardware 
or not?  If so, what impact did this have?

Thanks,

	- Godmar