[reactive] Bug fixes in progress

[Isaac Dupree]

Svein Ove Aas wrote:
> Certain people (*cough*quicksilver*cough*) thought it'd be nice of me
> to inform you of what I've been up to the last few days re: reactive
> bug-fixing, so here's an update.
> 
> - I've rewritten unamb, fixing all the issues I've managed to find
> (relating to nested unamb invocations, mostly) and having it throw a
> BothBottom exception if both values are a finite bottom.

Why not choose one of the two existing bottoms and throw that?  It'd be 
like the semantics of Haskell's imprecise exceptions anyway...

> I'm now trying to fix the joinE/>>=/mappend problems, but that might
> take a while. Meanwhile, more eyes on the rewritten code is better, so
> I've attached it; there hasn't been an official release yet.

okay I read it!  I just have some questions / suggestions to comment 
things better in the code, currently

> unamb = ...
>     where retry act = act `catch`

the definition of retry here makes almost no sense to me... but as I 
understand it, it's completely an RTS hack.  First I'd put a comment 
saying so, and then explain exactly what problem it avoids and how it 
does that, with more comments.

in "race": How does your current version kill descendent threads any 
better than "Simple version"?
Oh I see it puts the `finally` kill-the-threads in, around the 
read-the-mvar ("descentent" just means the two threads that "race" 
spawns, not the threads spawned by those two threads?).  But I'm not 
sure what the *extra* complication is for, that's beyond "Simple 
version"'s complexity?  Is it just in order to be able to throw 
BothBottom at the right time (a worthy goal)?  I also don't understand 
why putCatch wants to prevent certain kinds of exceptions from escaping 
the thread? (Would they emit into the calling-thread that `race` runs 
in? or what is the difference?  How did you choose the three or four 
things to catch; and what is wrong with NonTermination?)

thanks!

-Isaac