Re: [Caml-list] unix.select + camlp4

[skaller <skaller@users.sourceforge.net>]

On Mon, 2007-07-02 at 19:39 +0400, Anastasia Gornostaeva wrote:
> On Mon, Jul 02, 2007 at 12:10:34PM +1000, skaller wrote:
> > On Mon, 2007-07-02 at 01:01 +0400, Anastasia Gornostaeva wrote:
> > > On Sun, Jul 01, 2007 at 09:37:46PM +0100, Jon Harrop wrote:
> > 
> > > I dont understand: Try:
> > > 1) It receives "1-2+3*" and parses it. 
> > > 2) Later it receives "4" and continues parsing.
> > > 
> > > I want to get non-blocking parser. How to implement?
> > 
> > Put the parser in a thread, select in another thread.
> > Use Stream.from f for the parser. Write f to get characters
> > using the Event module, send the characters from the other thread
> > after select makes a channel active.
> 
> It looks awful if you want to run houndreds streams/connections.
> It is easier to run a connection with a parser in separate thread.
> So, camlp4 cannot be nonblocking parser.. bad.

This problem is not restricted to parsers .. it's a general
problem with Ocaml and also C, C++, and most other languages.

My language Felix solves this problem for procedures with
user space threading .. but it doesn't work with functions.
[And the builtin GLR parser is purely functional:]

Other languages like Scheme and I think Haskell and MLton
have more general solutions because they're not restricted
to the archaic concept of a linear stack.

You can certainly write code using continuation passing style
to work around this deficiency in most languages, but you
can't work around someone else's code that doesn't use this
style.

The general model of using 'threads' is the only solution,
however OS threads are too general and too slow for large
numbers, and they're needlessly asynchronous when all you
often want is control inversion (though in your case,
you're using 'select' so you do want asynchronicity).

JoCaml is worth looking at for another approach.. but the
bottom line is if you're using the usual parser products
around, the only way to control invert the 'get_token'
callback is by using a pair of threads implementing
a client/server model, and allowing the OS to control
invert the threads by stack swapping.


-- 
John Skaller <skaller at users dot sf dot net>
Felix, successor to C++: http://felix.sf.net