Re: [Caml-list] Re: Syntax extensions without Camlp4

[Alain Frisch <alain@frisch.fr>]

On 05/30/2012 04:14 PM, Hongbo Zhang wrote:
> 1. Why camlp4 is buggy?
> The main buggy part is its parsing technology.

I don't consider that the main problem with Camlp4 is that it is buggy, 
but rather that (i) it is overly complex for the benefits it delivers,
and (ii) it is actually not such a great idea to change the concrete syntax.

Some example of useless complexity:

  - A custom notion of AST. Why not simply use the OCaml one?  (Extended 
with nodes for a new nodes, like quotations.)

  - The use of concrete syntax for manipulating the AST.  The developer 
needs to understand not only the new AST, but also how it is reflected 
exactly by the concrete syntax quotations (and this is non trivial), and 
where anti-quotations are allowed, etc. What's wrong with normal pattern 
matching and expression building with the standard AST?  It might be a 
little bit more verbose, but it's so much simpler to understand.

  - A different syntax (the revised one).  I understand the benefits of 
this new syntax, but it seems kind of crazy to have a "low-level" tool 
implemented in (and encouraging) a syntax different from the core system.

  - A complicated bootstrapping cycle (partly a consequence of the fact 
that Camlp4 is itself written in a custom syntax).  That's mostly for 
OCaml maintainers, but in the past, it has slowed down development in a 
non-negligible way.


> 2. About the proposal.
> There are mainly 2 pieces. About the hook
> Parsetree.structure->Parsetree. structure, given that camlp4 already
> imports Parsetree, it's really trivial to
> add another hook after camlp4astdump2ocamlast.

I've absolutely no doubt that Camlp4 can be extended to be at least as 
powerful as this "Parsetree rewriting" proposal.  What's important is 
that this proposal is so simple that it can be implemented in a few 
dozens line of code in the core compiler. We should not create a 
dependency on a complex tool for problems which can be solved with 
something so simple.

> It's still nontrivial to write a robust Parsetree.structure ->
> Parsetree.structure, it would be nice if we could provide a quotation
> syntax for Parsetree.types.

I believe the opposite: it's simpler to write a robust AST->AST 
rewriting function if you work directly on the "real" AST definition, 
rather than a slighlty different one and with a custom syntax.  Just for 
an example, consider a left-hand side like:

   | <:pat< $p1$ | $p2$ | $p3$ >> -> ...

Will it capture both (p1|p2)|p3 and p1|(p2|p3)?  Or only one of them?
Another example: controlling precisely locations introduced in the AST 
fragments created with quotations is quite tricky.



Alain