[wg-camlp4] Negative field report: extension points unsuited for (matches p when e)

[Gabriel Scherer]

The recent caml-list message by rixed (archived below) motivated me to
get pa_matches in shape (it lacked support for "when ..."), and try to
see how a similar extension could be implemented with the new
extension_points branch.

The Camlp4 extension allows a new syntax for expressions, (matches
<pattern>) or (matches <pattern> when <expr>), that desugars into a
test function returning a boolean: (function <pattern> when <expr> ->
true | _ -> false). I have attached to this mail the relevant code,
called pa_matches.ml (pa_ is a traditional prefix for Camlp4 *parser*
extensions).

I quickly found out that extension points as currently implemented in
Alain's branch are ill-suited for this task, because the piece of code
we want to parse (<patt> when <expr>) is not syntactically an
expression. [%matches <patt> when <expr>] would be rejected by the
parser.

Note that this is just an experience report, and not a value judgment
on the extension_points approach itself. I think the choice to only
accept syntactically valid expressions (in fact structure items) has a
lot of merits (notably wrt. toolability). The price to pay is that
potential extension have to be massaged into syntactically correct
pieces, with a potential loss of concision when compared to
domain-specific syntax designs. This indicates, I think, that
extension_points is not always suited for "syntax changes in the
small", where a very concise syntax is given to an otherwise
expressible expression (and concision is thus essential for the
purpose of the extension), except when this extension exactly fits the
syntactic form of the more high-level OCaml syntactic classes.

For the purpose of experimentation, I still decided to do an
implementation based on a horrible hack, namely using (the new syntax
for) string literals that had been presented as a way to get "foreign
language quotations" in this context, with the syntax [%matches {|
<patt> when <expr> |}]. I believe this course of action should be
discouraged; in particular, there is no proper error handling. It
seems that deploying extensions based on this principle would
constitute a regression from the current Camlp4 situation.

Note that this implementation technique might be made slightly more
reasonable if the following ingredients were present:
- a simple and correct way to add location information from a lexing
buffer obtained from "Lexing.from_string", to inform the lexer of
where in the file it's picking up tokens (note that in the above case,
this is not enough as the quoted string is prefixed with some code to
make it a valid expression)
- more entry points in the parser for eg. patterns and match actions
- maybe support from the ppx wrapper to catch and re-route OCaml errors

I think this is all rather cumbersome and we should rather understand
and accept the limitations of extension points -- and make unruly
users respect them!

On Sat, Apr 27, 2013 at 10:04 AM, Gabriel Scherer
<gabriel.scherer at gmail.com> wrote:
> This is a subset of Richard Jones' "matches" syntax extension, turning
> an pattern matching on 'a into a ('a -> bool) function:
>
>   if (matches Some x when x > 15) variable then ...
>
> (The original extension did not handle "when", but it would be quite
> easy to add.)
>
> On Sat, Apr 27, 2013 at 9:18 AM,  <rixed at happyleptic.org> wrote:
>> Wouldnt it be nice if, as a special form of the = operator,
>> "variable = pattern when condition" were equivalent to
>> "match variable with pattern when condition -> true | _ -> false" ?
>>
>> So that one could easily check optional values for instance:
>>
>> if variable = Some x when x > 15 then blabla
>>
>> What do you think?
>>
>>
>> --
>> Caml-list mailing list.  Subscription management and archives:
>> https://sympa.inria.fr/sympa/arc/caml-list
>> Beginner's list: http://groups.yahoo.com/group/ocaml_beginners
>> Bug reports: http://caml.inria.fr/bin/caml-bugs
-------------- next part --------------
A non-text attachment was scrubbed...
Name: pa_matches.ml
Type: application/octet-stream
Size: 1914 bytes
Desc: not available
URL: <http://lists.ocaml.org/pipermail/wg-camlp4/attachments/20130501/66c3b30b/attachment.obj>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: ppx_matches.ml
Type: application/octet-stream
Size: 1208 bytes
Desc: not available
URL: <http://lists.ocaml.org/pipermail/wg-camlp4/attachments/20130501/66c3b30b/attachment-0001.obj>