Re: [Caml-list] gc question: thread stacks, fibers, etc.

[Chris Hecker <checker@d6.com>]

>An alternative is to allocate the stack outside the heap, and scan it
>via scanning hooks like the thread library does, but manipulate it
>from the Caml side through a proxy, heap-allocated custom block.  The
>custom block has a finalization function that the GC will call when no
>reference to the proxy (i.e. to the fiber) remain.  The finalization
>function can then decide to kill the fiber (free its stack) if it so
>pleases.

But this doesn't solve the original problem of the fiber containing a 
reference to itself on its stack (or any fiber points to any other fiber), 
right?  The scanning hook functions assume they're scanning roots, so they 
can't check for circularity and release the whole mess.  The function I 
need is something that says "the values I'm sending to the scanner are not 
stored in roots, so check for circularities that includes them".  Or 
something like that.  Make sense?  Perhaps there's a way to hook that in 
the gc?  Maybe I'm missing something.

Anyway, ignoring the circularity problem for the moment, deleting the stack 
from under a fiber works because the stack itself can never have a pointer 
from the heap into it, so it's always okay to delete (according to your 
answer to my other question).  I guess you'd have to be careful that no C 
functions that were still on the stack of that fiber had ever passed a 
pointer into their stack to anyone who keeps it elsewhere (which in 
ordinary cases is okay, since when writing the C code you assume you'll get 
returned-to in an orderly fashion).  Maybe that's an okay constraint to 
place on code, but a bug of that nature would certainly be hard to find, so 
maybe it's not okay (this is why people shouldn't kill threads, nb. the 
other thread  :).

Hmm, now that I think about it, I can just use an exception for this, 
exactly like the other caml-list discussion was wishing it could for 
threads.  In other words, I'm only ever going to consider toasting a fiber 
when it's yielded, so when the fiber gets resumed it'll be in the yield 
code.  That code can check whether the gc has requested finalization of the 
fiber, and throw an exception back up through the fiber.  That will solve 
all the stack issues, because the exception will unwind it all, and the 
fiber can catch it if it wants and do any cleanup.

Yes, that's much cleaner than just killing it anyway.  That just leaves the 
circularity issue above.

Thanks!
Chris

-------------------
To unsubscribe, mail caml-list-request@inria.fr Archives: http://caml.inria.fr
Bug reports: http://caml.inria.fr/bin/caml-bugs FAQ: http://caml.inria.fr/FAQ/
Beginner's list: http://groups.yahoo.com/group/ocaml_beginners