On Mon, 2007-09-24 at 17:23 +0200, Nicolas Pouillard wrote:

> > (* My Ocaml File .. *)
> > (*#Syntax Mycamlp4syn *) (* for compatibility *)
> > #syntax Mycamlp4syn;     (* better *)
> 
> This  sounds  like  a  good  idea.  That's  not a new one and it was our first
> choice  when  designing  ocamlbuild. However after some reflections we decided
> that  one  can  first do _tags method and then provide a way to read tags from
> the file itself.

There's slight difference in view as expressed here. 
I wouldn't suggested arbitrary "tags" or hints be first class syntax
(although one might put them in comments).

I'm not against tags annotations.. I'm suggesting the specification
of the syntax extension doesn't belong on the command line, 
or in a tags file, it belongs in the file.

> The  main  problem  with  the  solution with embedded tags is that one must at
> least  define  a  how to read these annotations. This is not that easy since a
> .ml  file  can  be  a  pre-processed file with a totally different syntax. 

No, this should not be allowed: that's my point. Every Ocaml file uses
THE syntax for either *.mli or *.ml, end of story. If it uses some
other language, it should have a different extension.

However the syntax can be extensible, but the extensions must be 
at least loaded, if not defined, IN the file.

Felix does this. It uses Dypgen GLR parser and OCS Scheme
to implement it. You have to say:

	open syntax felix;

or you can't parse anything except syntax extensions :)
The whole of the language proper (except nonterminals
mapped to terminals such as literals) is specified
by a dynamically specified grammar. In Felix,
you can extend the grammar 'anywhere', the extensions are
properly scoped.

So actually, there's no problem designing a syntax which allows
loading another syntax, that is, a bootstrap syntax. For Ocaml,
you could easily add a directive

	open syntax Mysyntax

or

	#syntax Mysyntax

to the standard grammar, and make camlp4 parse it and use it
to load syntax extensions. Running camlp4 so this happens should
be the mandatory, no -pp command stuff: camlp4 IS the Ocaml parser.

Then, when you run, say, ocamldoc, it knows which extensions to load,
as does the human reader.

It's just the same as

	let (@) f g x = f (g x)

The "syntax extension" of the user defined operator @ for composition
is defined IN the program.

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