Re: Ramifications of Removing Coroutines or Implement at higher level than C

Re:  Coroutines compile option -

Oh there's a USE_VALGRIND option?  Sweet.

Re:  Catch C++ exceptions at binding site -

I do already catch all C++ exceptions in the bindings.  However, my C++ and Io interaction is highly re-entrant.  This is what my call stack looks like:
 - C++ launches Io
 - Io calls a C++ function
 - The C++ function activates an IoBlock
 - Io code in the block may attempt to go to another coroutine
 - Io code on the other coroutine may also call C++ functions
 - When this nested C++ code throws an exception, the process wigs out *BEFORE* I even get to my catch block around the binding site.  In fact it even crashes the process to try throw and catch in one line of code:  try { throw exception; } catch (exception) { never gets to break point here };.

It's still possible this is due to some problem in my C++ code (like a bad memory free in a destructor).  But I think the C++ exception mechanism itself is becoming corrupted.  I haven't done x86 assembly programming in a long time but I guess I'm going to have to debug this one at that level to really know for sure.

Re:  Just change the implementation of coroutines, not remove them -

This is what I really mean.  Yeah I'd probably do it with threads and a global interpreter lock.  Not pretty but shouldn't require very much code change.  Or do it at the interpreter level like Stackless Python, but that would be a much larger change.  I might do that if I eventually port Io to C++ as I've thought off and on about doing.  (If I did that it would be a friendly gesture to C based Io, saying "Io is so cool it deserves more than one implementation", not "I'm in a huff so I'm going to take my ball and play somewhere else.")  In that case I wouldn't be worried about losing support for Exceptions because I would just use C++ exceptions.


--- In [email protected], Steve Dekorte <steve@...> wrote:
>
> 
> On C++ exceptions hosing coros: 
> 
> 	Wouldn't putting C++ catch on every location that Io call's into C++ in your bindings resolve it?
> 
> On valgrind:
> 
> 	Did you define USE_VALGRIND so Coro.c will handle it?
>