[LLVMdev] RFC: Exception Handling Proposal Revised

[Renato Golin]

On 1 December 2010 20:02, Duncan Sands <baldrick at free.fr> wrote:
> thanks for the explanation.  I am writing up an alternative exception handling
> proposal which does not suffer from this problem (at least I think it doesn't!).

Hi Duncan,

Remember that a big reason why to change was to support instructions
that could throw exceptions (such as divide in Java). The idea of
having all those dispatch/landingpad was to ultimately get rid of the
invoke instruction and move the information to the basic block, so
every instruction (including function calls) could, in theory, throw
an exception.

In this case, it'd be easier for the back-end to generate the EH table
regions, as it wouldn't need to consider invokes, only groups of basic
blocks (Bill's regions).

cheers,
--renato