OCaml Weekly News Previous week Up Next week

Hello

Here is the latest OCaml Weekly News, for the week of January 13 to 20, 2015.

  1. ppx_monadic.1.0.2, ppx for monadic do, pattern guards and monadic comprehension
  2. Quizz
  3. ppx_deriving 2.0, ppx_deriving_yojson 2.3
  4. Detecting dead code, discovering a project's structure...
  5. ocamlnet-4.0.1
  6. new releases: pxp, webdav, cache, netamqp, wdialog
  7. Other OCaml News

ppx_monadic.1.0.2, ppx for monadic do, pattern guards and monadic comprehension

Archive: https://sympa.inria.fr/sympa/arc/caml-list/2015-01/msg00061.html

Jun Furuse announced:
I have OPAM-released ppx_monadic.1.0.2, a ppx extension for

* Monadic do notation:  do_;
* Pattern guards:  when p <-- e -> ..
* List and monadic comprehension:  [%comp x || p <-- e; ..]

These 3 extensions share the same do-sequence syntax sugar, which is a
sequence of the following phrases:

* Monadic bind : p <-- e
* Action: e
* Escape: (); e    (to write the normal sequential execution)
* let .. in .. and [%x .. ]

It is already available via OPAM: opam intall ppx_monadic. More
detailed documentation and source code is available at:
https://bitbucket.org/camlspotter/ppx_monadic

Enjoy!
      
oleg then said and Jun Furuse replied:
>> * Monadic do notation:  do_;
>> * Pattern guards:  when p <-- e -> ..
>
> This reminds me: MetaOCaml incorporates the let! notation suggested by
> Nicolas Pouillard and implemented by Alain Frisch. The following is
> the example of using the notation, excerpted from the file
>         metalib/test/pythagorian_triples.ml
> in the BER MetaOCaml distribution/source.
>
> (* The example uses left recursion and truly infinite streams! *)
> (* Don't try this in Prolog or in Haskell's MonadPlus. *)
>
> let rec numb () =                       (* infinite stream of integers *)
>     yield (mplus (let! n = numb in ret (n+1))         (* left recursion! *)
>                (ret 0)) ()
> ;;
>
> let pyth : (int * int * int) NonDet.stream =
>   let! i  = numb in
>   let! () = guard (i>0) in
>   let! j  = numb in
>   let! () = guard (j>0) in
>   let! k  = numb in
>   let! () = guard (k>0) in
>   (* Just to illustrate the `let' form within let! *)
>   let test x = x*x = j*j + k*k in
>   let! () = guard (test i) in
>   ret (i,j,k)
> ;;
>
> let [(5, 4, 3); (5, 3, 4); (10, 8, 6); (10, 6, 8); (13, 12, 5); (13, 5, 12);
>      (15, 12, 9); (15, 9, 12); (17, 15, 8); (17, 8, 15)]
>  =
> run 10 pyth;;
>
>
> (The file also implements the non-determinism monad that is
> illustrated in the example.)

Thanks,

let! is not valid in the vanilla syntax therefore I guess you have
patched the compiler.

Current OCaml syntax has shortage of pattern binding expression and
only usable is let%xxx p = e in which is a bit pain for ppx writers.

The pattern part of   p <-- e  is actually an expression and you
cannot simply write real patterns like   (x, y) as z. You still can
write it using an extension like [%p? (x,y) as z] but it is lousy...
      
oleg then said and Alain Frisch replied:
>> Current OCaml syntax has shortage of pattern binding expression and
>> only usable is let%xxx p = e in which is a bit pain for ppx writers.
>
> Indeed. One may wish that
>
>          let rec p = e1 in e2
>
> were treated as if it were
>          let[@ocaml.let "rec"] p = e1 in e2
>
> and likewise let module.

Please, no!  Attributes are intended to add meta-data for external
tools (ppx, tools parsing .cmt files, etc), perhaps also to tweak the
behavior of the compiler (trigger/control warnings, etc), certainly
not to encode core language features (otherwise, let's use
s-expressions instead of Parsetree).  Facilitating language
experiments is also a good use for attributes, but not as a long-term
solution for the official compiler.

> Come to think of it, 'rec! or 'module' are
> annotations on let.

"let module" has a different shape (module
identifiers/types/expressions) than "let".  And it doesn't seem
realistic to merge, say, core types and module types, or core
expressions and module expressions.
      
Gabriel Scherer then said and Alain Frisch replied:
> (On the other hand, the remark that the existing uses of keyword-bang in
> the language, namely (method!) and (open!), could now be represented as
> annotations is fairly reasonable.)

Indeed, these ! markers only affect warnings, and don't change the
static or dynamic semantics otherwise, so I wouldn't be shocked if
they were represented internally with attributes.  Unless perhaps if
we consider that forcing users to mark method overrides, as in many
other class-based OO languages, should become part of the language
definition (i.e. turned into a strong error at some point).
      
Jeremy Yallop then added:
A nit: for methods, at least, the ! already produces an error in the
case where the method is not previously defined:

   # object method! m = 3 end;;
  Characters 7-20:
    object method! m = 3 end;;
           ^^^^^^^^^^^^^
  Error: The method `m' has no previous definition
      
Alain Frisch also replied to Jun Furuse:
> Current OCaml syntax has shortage of pattern binding expression and
> only usable is let%xxx p = e in which is a bit pain for ppx writers.

If it's a pain, it's rather for users of ppx, no?

Providing "alternative" forms of existing constructs (i.e. forms that
look syntactically similar to these constructs) to be interpreted
before they reach the type-checker, is indeed what extension nodes
[%foo ...] were designed for.  For expression constructs starting with
a keyword, the short form KEYWORD%foo ... makes the use of extension
nodes very cheap syntactically.

I can appreciate that authors of tools that requires special syntactic
support would love to have their new forms look completely native to
users, but the counter-argument can be made that keeping an explicit
syntax (through the '%' character) for features that are not part of
the official language is a good property.  (Camlp4/campl5 are still
available for people who want to play with the concrete syntax.)


With the short form, an alternative binding construct can be as short as:

 let%L p = e in ....

I'm not suggesting to reserve one-letter extension identifiers for
specific projects, but each tool could let the user choose explicitly
which letter to use:


  [%%Project1 alt_id = "L"]
  [%%Project2 alt_id = "A"]
  ...
  ...
    let%L p1 = e1 in ....
    let%A p2 = e2 in ....
      
Drup then said and Jun Furuse replied:
> I personally like the explicitness of the syntax a lot. The only issue
> in OCaml currently is that, given the need for retro compatibility, it
> goes sometimes against the terseness. For example the impossibility to
> do " x@foo" instead of "x[@foo]". That's unavoidable, though.
> 
> This is, by the way, a point I dislike a lot with ppx_monadic. It
> abuses the native syntax in completely alien ways and without
> annotations.
> 
> I like ppx_monad's syntax quite better due to the fact that it's
> always explicitly a ppx (thanks to %monad) and do not overload the
> "do_" identifier.

I know what you wrote, since I always asked myself “isn’t it too
bizarre?” when I was writing ppx_monadic. Here are some justifications
I made:

The first intension of ppx_monadic is to port my code with pa_monad
without pain. Unfortunately ppx_monad required too many small fixes
due to the following reasons:

· “perform” needed to be replaced by “begin%monad .. end” or “[%monad .
. ]” which requires closing, which is really pain. Ppx_monad provides
“fun%monad”, “match%monad” without closing but they were not always
helpful for me.
· The pattern “p” of “p <- e” is limited to a variable, since it
overrides the syntax of object member mutation. In addition, I use
lots of monadic binds inside my class methods. Therefore this override
is not acceptable for me.

Secondary, I see the monadic syntax sugar is to reduce the number of
key types, and I accepted some alien syntax weirdness for the purpose.
If the number of key types would not matter, I would be happy with the
good old bind (>>=) chains and would not use “do” at all. People think
bind chains are hard to read but my experience with Jane Street Async
tells me it it not quite. Ppx_monad does not really gain in this point
unfortunately: I was often forced to write “let%monad (x,y) = e in”
just for “(x,y) <-- e;”. I write Haskell do-notations daily and wanted
to have something comparable in OCaml.

Anyway, the current OCaml syntax is limited to have do-notation which
makes everyone happy. If there would be a way in OCaml to write “foo
e” for some keyword “foo” which is like “begin” but does not require
“end”, I would be pretty happy to change the weird “Option.do_; e” to
“foo%Option e”. Before implementing “do_; e”, I tried a bit of “[%do]
e” but it did not work well since “[%do] p <-- e” is parsed as “([%do]
p) <-- e”, not “[%do] (p <-- e)”.
      

Quizz

Archive: https://sympa.inria.fr/sympa/arc/caml-list/2015-01/msg00064.html

Thomas Braibant asked:
Here is a quizz that can easily be solved using OCaml (version
4.02.1), but it would be less fun than trying to infer which functions
are accepted.

```
module type S = sig val test : 'a list end

type t = (module S)

let a (x : t) =
  let module M = (val x) in
  List.length M.test

let b (module M : S) =
  List.length M.test

let c ((module M) : t) =
  List.length M.test

let d (module M : t) =
  List.length M.test

let e (x : (module S)) =
  let module M = (val x) in
  List.length M.test

let f ((module M) : (module S)) =
  List.length M.test

let g : (module S) -> int = fun (module M) ->
  List.length M.test

let h ((module M) : (module sig val test : 'a list end)) =
  List.length M.test

```
(I am sure I have missed interesting variations, though.)

I have read through

http://caml.inria.fr/pub/docs/manual-ocaml/types.html
http://caml.inria.fr/pub/docs/manual-ocaml/modtypes.html
http://caml.inria.fr/pub/docs/manual-ocaml/extn.html#sec230

but what I do not get is how to make a variation of `h` work. That is,
I would like to inline the module signature. Is that possible?
      
Leo White replied:
In OCaml 4.01 and earlier, first-class modules were typed
nominatively. In other words, their type was based on the name of the
module type used, not its structure. For example:

            OCaml version 4.01.0

    # module type T = sig type t end;;
    module type T = sig type t end

    # module type S = sig type t end;;
    module type S = sig type t end

    # let f (x : (module T)) : (module S) = x;;
    Characters 38-39:
      let f (x : (module T)) : (module S) = x;;
                                            ^
    Error: This expression has type (module T)
           but an expression was expected of type (module S)

This obviously requires that the module type has a name:

    # type t = (module sig type t end);;
    Characters 17-20:
      type t = (module sig type t end);;
                       ^^^
    Error: Syntax error

In version 4.02, the type-checking of first-class modules was changed to
be structural instead of nominative:

            OCaml version 4.02.1

    # module type T = sig type t end;;
    module type T = sig type t end

    # module type S = sig type t end;;
    module type S = sig type t end

    # let f (x : (module T)) : (module S) = x;;
    val f : (module T) -> (module S) = <fun>

This means that it would now be possible to support first-class modules
for types which have no name, but such support has not yet been
added. (There are some awkward issues around the syntax in relation to
with constraints, but they are probably solvable.)

So, in summary, you can't do that, but it is now at least theoretically
something you could do.
      
Milan Stanojević then asked and Leo White replied:
> Are there any concrete plans to add syntax support?

I don't think there are any concrete plans. Jacques mentioned it was a
possibility in a comment on Mantis:

  http://caml.inria.fr/mantis/view.php?id=6333#c10990

and other than that I don't think anyone has mentioned it.
      
Jacques Garrigue replied:
The main problem is not so much syntax, as the fact it would require to make
all definitions in the Types module mutually recursive. Not only that, but
operations like path substitution need to be mutually recursive in the same
way. So the question is whether the small gain in flexibility is worth making the
implementation more complex.
(Note that an extra gain is that it becomes possible to expand a module type
definition when leaving its scope)
      
Milan Stanojević then asked, Leo White replied, and Jacques Garrigue confirmed:
>> In my work I mostly just wanted to be able to just do something like
>> (module M : Intable with type t = t), i.e just specializing existing
>> module type with "with type =" or "with type :=".
>> Is this special case any easier to implement?
> 
> "with type t =" already works. I'm not sure, but I think that "with type
> t :=" could indeed be implemented without the increased implementation
> complexity that Jacques was referring to.

Indeed it could.
Module type names + non-parameterized constraints are ok, because you only have
to recurse on types, not concrete signatures.
      

ppx_deriving 2.0, ppx_deriving_yojson 2.3

Archive: https://sympa.inria.fr/sympa/arc/caml-list/2015-01/msg00080.html

Peter Zotov announced:
I'm glad to announce the releases of ppx_deriving[1] and
ppx_deriving_yojson[2]. These releases add support
for open types. They will be available in OPAM shortly.

Many thanks to Zoggy for implementing this feature.

[1]: https://github.com/whitequark/ppx_deriving
[2]: https://github.com/whitequark/ppx_deriving_yojson
      

Detecting dead code, discovering a project's structure...

Archive: https://sympa.inria.fr/sympa/arc/caml-list/2015-01/msg00081.html

Yoann Padioleau replied to this old thread:
Indeed the codegraph tool, part of Pfff, does some
global analysis of an ocaml project using the .Mt (generated by ocamlc
-bin_annot) and generates
an index of a codebase (a graph_code.marshall file).
Not all features of Ocaml are supported though (e.g. functors).

Once the graph_code.marshall file has been generated,
you can use check, also part of pfff, to detect dead code
or other errors.

https://github.com/facebook/pfff/wiki/CodeGraph
https://github.com/facebook/pfff/wiki/Scheck
      

ocamlnet-4.0.1

Archive: https://sympa.inria.fr/sympa/arc/caml-list/2015-01/msg00087.html

Gerd Stolpmann announced:
after some testing I could finally release OCamlnet-4.0.1. Find it at
the usual place:

http://projects.camlcity.org/projects/ocamlnet.html

This is a major release, and there are incompatible changes, in
particular in the following areas:

 - How ocamlnet is configured (new switches: -enable-gnutls,
   -enable-gssapi, gone: -enable-ssl, -enable-crypto)
 - How to use TLS (formerly SSL) in protocols (see below)
 - Unicode tables need to be explicitly linked in (package
   netunidata), or conversions between character encodings
   may be unavailable
 - IPv6 is now enabled by default if there is a global IPv6
   address
 - A number of modules has been renamed (e.g. Http_client is
   now Nethttp_client)
 - A number of modules have been split up (e.g. some functions
   of Uq_engines are now in Uq_client and other modules)

But these are only the most visible changes. Besides that, there are
also plenty of new features, all about authentication and security.

1. TLS

The old support for OpenSSL (via the ocaml ssl bindings) are no longer
supported. Instead, there is an integrated wrapper for GnuTLS, which has
much nicer API and has better support for the latest TLS updates.

Read more here:
http://projects.camlcity.org/projects/dl/ocamlnet-4.0.1/doc/html-main/Tls.html

2. SASL authentication

OCamlnet-4 implements all important SASL authentication mechanisms (for
both clients and servers): PLAIN, CRAM-MD5, DIGEST-MD5, SCRAM-SHA1,
GS2-KRB5 (more on that one below). SASL is integrated into the protocol
interpreters for POP and  SMTP.

3. GSSAPI/Kerberos

Before this release, the was only very preliminary support for the
GSSAPI. This has changed, as there is now a binding of the system-level
GSSAPI functions. In particular, this permits protocol interpreters to
support Kerberos, and in Ocamlnet-4 SunRPC, HTTP clients, and FTP have
been "kerberized" this way. You can also run Kerberos on top of SASL via
bridging.

If you are unfamiliar with Kerberos, this is a single-sign-on security
framework for LANs, and it can be frequently found in company networks.

At the moment, we cannot support Kerberos on Windows, though (as Windows
doesn't provide a standard-compliant GSSAPI library).

4. Pluggable security modules

The security mechanisms are all pluggable. If you e.g. want to use a
different TLS provider, you "only" need to ensure that it has a module
interface compatible with TLS_PROVIDER. There's also SASL_MECHANISM,
HTTP_MECHANISM, and GSSAPI. The mechanisms are passed as normal values
to the protocol interpreters thanks to first-class modules. This permits
it to support several implementations, and also to weakly bind to
security mechanisms (e.g. use only TLS if available).
      

new releases: pxp, webdav, cache, netamqp, wdialog

Archive: https://sympa.inria.fr/sympa/arc/caml-list/2015-01/msg00088.html

Gerd Stolpmann announced:
as followup to OCamlnet-4.0.1 there are also new releases of these
libraries that have been ported to OCamlnet-4.0.1:

http://projects.camlcity.org/projects/pxp.html
http://projects.camlcity.org/projects/webdav.html
http://projects.camlcity.org/projects/cache.html
http://projects.camlcity.org/projects/netamqp.html
http://projects.camlcity.org/projects/wdialog.html
      

Other OCaml News

From the ocamlcore planet blog:
Thanks to Alp Mestan, we now include in the OCaml Weekly News the links to the
recent posts from the ocamlcore planet blog at http://planet.ocaml.org/.

Unikernels for everyone!:
  http://amirchaudhry.com/unikernels-for-everyone

Work summary @ OCaml labs:
  http://yansnotes.blogspot.com/2015/01/work-summary-ocaml-labs.html
      

Old cwn

If you happen to miss a CWN, you can send me a message and I'll mail it to you, or go take a look at the archive or the RSS feed of the archives.

If you also wish to receive it every week by mail, you may subscribe online.


Alan Schmitt