Re: Coroutines and blocking IO

[Javier Guerra <javier@...>]

On Thursday 26 January 2006 12:58 am, Jon Smirl wrote:
> > keplerproject.org uses copas.  IIRC, you can create a copas socket that
> > wraps a luaSocket, but has the same interface (at least the same
> > interface needed after you wrap it).
> >
> > If Copas doesn't actually do this, it would be extremely easy to add it
> > in.

it's in Copas, not Xavante itself.  but the wrapped API is a small subset of 
luaSocket's, just what we needed for Xavante.  (but it's very easy to extend 
it)

> Wrapping of blocking functions is key to how Ruby implements threads.
> Green threads don't work if an application thread can get stuck in a
> blocking system call. By wrapping everything in the run-time Ruby
> makes sure you can't get stuck.

sounds a nice feature; but there's one point most people seems to be missing: 
Lua has no real threads (green or whatever).  Coroutines are not threads.

of course, given the power (coroutines), we build the mechanism (threads); 
that's what Copas is about.  it works, but it's far from a really complete 
threading mechanism.

i've glanced around Diego's dispatch.lua (in the latest LuaSocket); and it 
seems to be far simpler and cleaner than Copas.  at least, the wrapped API is 
much closer to the full LuaSocket API.

but i feel that most of us wouldn't like to get a wrapping mechanism written 
in stone (in the language itself).  much like there's no fixed and standard 
OO system in Lua.  (of course, from time to time someone asks for one... but 
most of us just implement one when needed)

what i'd like to see is the LuaSocket dispatch system extended a bit more, to 
make it very easy for other module writers to create non-blocking calls that 
cooperate with the dispatch.  that way, we would have a de-facto threading 
(and wrapping) system.

also, i'd like to see non-blocking file IO (and database, and process 
spawning, and...) all pluggable into dispatch.lua

-- 
Javier

Attachment: pgp8lx01ZtWPh.pgp
Description: PGP signature