llvm dev - Jun 2011 - [LLVMdev] Is LLVM expressive enough to represent asynchronous exceptions?

On 14 June 2011 02:03, Bill Wendling <wendling at apple.com>
wrote:
> You laugh, but we support it (or at least we did at one point with
gcc-4.2).
In IR? In the form of synchronous exceptions?

How can you possible represent a "signal cfg" in IR? Is there a need
for unwinding information at all?

cheers,
--renato

On Jun 14, 2011, at 1:37 AM, Renato Golin wrote:
> On 14 June 2011 02:03, Bill Wendling <wendling at apple.com> wrote:
>> You laugh, but we support it (or at least we did at one point with
gcc-4.2).
> 
> In IR? In the form of synchronous exceptions?
> 
> How can you possible represent a "signal cfg" in IR? Is there a
need
> for unwinding information at all?
> 
My understanding was that we support exceptions thrown from signals. The
mechanism for modeling this is unknown to me (it's most likely a gcc thing
and not an llvm thing).

-bw

On 14 June 2011 19:07, Bill Wendling <wendling at apple.com>
wrote:
> My understanding was that we support exceptions thrown from signals. The
mechanism for modeling this is unknown to me (it's most likely a gcc thing
and not an llvm thing).
Oh, I see. That's a slightly different matter, but still, unwinding
makes little sense when the signal (and thus its exception) can occur
at any time meaning anything.

This does look like a very specific feature for a very specific use
case (which I can't fathom). ;) But I don't think that will be
considered on a possible refactoring of the exception handling in
LLVM.

thanks,
--renato