Hello
Here is the latest Caml Weekly News, for the week of 22 to 29 March, 2005.
Archive: http://caml.inria.fr/pub/ml-archives/caml-list/2005/03/880ebc2e6c2a645a0ffb55df3cb79e2d.en.html
Frédéric Gava asked:
I have a problem with an initialization of a array of strings using a C
procedure. I give you the code
with some comments. My code works well without threads but when I add some
threads to my code the array is initializing with some strange strings. If
someone could help me...
Thanks,
Frédéric Gava
char **argv;
/* "arguments" is the Ocaml Sys.argv array */
value bsmlpub_init(value arguments)
{
int argc,i;
CAMLparam1(arguments);
/* alloc the arguments */
argc = Wosize_val(arguments);
argv = (char**)stat_alloc((argc + 1) * sizeof(char *));
for (i = 0; i < argc; i++) argv[i] = String_val(Field(arguments, i));
argv[i] = NULL;
/* saving them */
/* This function is a C function take from a special library for parallel
computing. So it is impossible to modified this function. This function
gives to all processors the same parameters take from the line command of
the bash. */
bsplib_saveargs(&argc, &argv);
CAMLreturn (Val_unit);
}
/* Them, when a program wants to read the parameters, it calls this
functions, but here
I have an array with bad strings.... */
value bsmlpub_argv(value unit)
{
CAMLparam1(unit);
CAMLreturn (copy_string_array(argv));
}
Olivier Andrieu answered:
using String_val to pass a caml string to a C function can be unsafe
in several cases.
Here I guess the C function copies the char* pointers and stores then
somewhere. When they are accessed later (by another function), they
have become invalid because the GC has moved the strings around. Of
course you can't really know what the function does with the pointers
from the prototype alone, you'd need to see the library source code
(or a good documentation). The solution would be to copy the strings with
strdup() :
for (i = 0; i < argc; i++) argv[i] = strdup (String_val(Field(arguments, i)));
Archive: http://caml.inria.fr/pub/ml-archives/caml-list/2005/03/d9e38d879617f83ed6630edbf8ab9a2b.en.html
Markus Mottl described:
though the idea is not new, more precisely, I have shamelessly stolen
it from Gerd Stolpmann's findlib sources, it still seems not widespread
and is such a useful thing to have that it should be mentioned here on
the list.
You will quite often want to have custom pretty-printers for your
datatypes in the toplevel. Preferably, they should be installed as
soon as the library implementing them has been loaded.
The solution for this is quite simple, and I have just implemented it
in the LACAML-library so that matrices and vectors can be printed in
the toplevel.
Assuming you have already implemented a pretty-printer with the correct
signature in module "My_module":
val pp_my_type : Format.formatter -> my_type -> unit
Then just create another module in your library:
---------------------------------------------------------------------------
let printers =
[
"My_module.pp_my_type";
]
let eval_string
?(print_outcome = false) ?(err_formatter = Format.err_formatter) str =
let lexbuf = Lexing.from_string str in
let phrase = !Toploop.parse_toplevel_phrase lexbuf in
Toploop.execute_phrase print_outcome err_formatter phrase
let rec install_printers = function
| [] -> true
| printer :: printers ->
let cmd = Printf.sprintf "#install_printer %s;;" printer in
eval_string cmd && install_printers printers
let () =
if not (install_printers printers) then
Format.eprintf "Problem installing printers@."
---------------------------------------------------------------------------
Just add further pretty-printers to the list "printers" if required.
If you link this module into a byte-code library and load it into the
toplevel (using #load), you will be able to print your types using the
custom pretty-printers.
There is one caveat here: the resulting library cannot be used for linking
with byte-code executables, because the toplevel is not available there.
Therefore, if you want to implement a library, you will have to link
two separate ones: one with and one without the module for installing
the printers.
If you happen to use "ocamlfind" to "#require" your libraries, you just
need to add a line to the META-file for each kind of library. The right
one will be chosen automatically then, e.g.:
archive(byte)="mylib.cma"
archive(byte,toploop)="mylib_top.cma"
Here is a short demonstration using the pretty-printers of the
LACAML-library in the toplevel while computing the singular value
decomposition of a 3x3 Hilbert-matrix:
---------------------------------------------------------------------------
# #require "lacaml";;
/usr/local/home/godi/godi/lib/ocaml/std-lib/bigarray.cma: loaded
/usr/local/home/godi/godi/lib/ocaml/site-lib/lacaml: added to search path
/usr/local/home/godi/godi/lib/ocaml/site-lib/lacaml/lacaml_top.cma: loaded
# open Lacaml.D;;
# let mat = Mat.hilbert 3;;
val mat : Lacaml_float64.mat =
1 0.5 0.333333
0.5 0.333333 0.25
0.333333 0.25 0.2
# gesvd mat;;
- : Lacaml_float64.vec * Lacaml_float64.mat * Lacaml_float64.mat =
( 1.40832
0.122327
0.00268734,
-0.827045 0.547448 0.127659
-0.459864 -0.52829 -0.713747
-0.323298 -0.649007 0.688672,
-0.827045 -0.459864 -0.323298
0.547448 -0.52829 -0.649007
0.127659 -0.713747 0.688672)
---------------------------------------------------------------------------
Enjoy this trick...
Archive: http://caml.inria.fr/pub/ml-archives/caml-list/2005/03/3cc2685923f947427957693d6b2027ce.en.html
Jeffrey Cook asked:
Is there an existing (or even possible to do with) camlp4 extension to
support default values in records? I am constantly finding myself
doing the following:
type bob = {
foo : string ; <--- field for which no default value makes sense
bar : int ; <--- field(s) that have default values
}
let bob_default = { foo = ""; bar = 0; }
with usage:
let x = { bob_default with foo = "hello" }
so that I do not have to fill in the numerous fields that really
should default to some value.
Without using this solution, my code is often cluttered with many
doldrum default value declarations for record fields (worsened by
multiple sites that define the initial record values), obscuring what
the important assignments were. Additionally, I often find that the
complex fields (of record types, for example) are the ones that don't
lend themselves to a default value whereas the simple types do. This
leads to the use of options (discussed below) just to save from even
uglier default value declarations for fields that never use a default.
Using a default structure like above also does not let me harness the
type checking mechanisms to detect places in my code where a field
(that really shouldn't have a default value) wasn't defined as is
normally the case when constructing a record from scratch - especially
when fields are later added to the record.
Using 'options' for no-default-value fields doesn't seem to be a very
good solution, as there then need to be runtime deconstructors of the
options (and runtime checks or uncaught exceptions) if a field has a
None value that should have been set. All that aside, it still
wouldn't identify where the record was incompletely 'created' using
the default record.
What I would like is the following:
type bob = {
foo : string ;
bar : int := 0 ;
}
let x = { foo = "hello" }
(and maybe even some kind of 'nodefaults' annotation when defining 'x'
if I want to make sure I've explicitly defined every field.)
So what advice / solutions do other people have / use?
Martin Jambon answered:
I would go for a "create_bob" function with labelled arguments, some of
them being optional:
let rec create_bob ~foo ?(bar = 0) () =
{ foo = foo;
bar = bar }
Implementing a syntax extension that defines "create_bob" automatically is
typically something that can be done with Camlp4 :-)
It supports recursive type definitions such as:
type t = { x : t option = Some { x = None } }
or even (since create_t is recursive):
type t = { x : t option = Some (create_t ~x:None ()) }
which would be expanded into:
type t = { x : t option }
let rec create_t ?(x = Some (create_t ~x:None ())) () = { x = x }
Archive: http://caml.inria.fr/pub/ml-archives/caml-list/2005/03/47367327f0a6020488a35e56d397aae1.en.html
Pau Giner announced:
Hello, I designed some time ago a new logo proposal for O'caml
(http://usuarios.lycos.es/DrNotrix/images/ocaml.png). Now I've made a
new version more suitable for buttons (caml powered like) or icons.
You can find it at:
http://usuarios.lycos.es/DrNotrix/images/ocaml-icon.png
The design represents a *camel* (a bit *abstract*) but it's also a
*machine*. So I think it fits with the concept of the language and
can't be confussed with perl logo.
Your feedback is welcome.
Alex Baretta said:
Actually, the Perl logo is a *dromedary*. Ours is supposed to be Cam(e)l.
Archive: http://caml.inria.fr/pub/ml-archives/caml-list/2005/03/99c702fe92423a1bc18c92a53d7bc010.en.html
Alex Baretta asked, Kip Macy said, and Gerd Stolpmann answered:
Kip Macy wrote:
> Alex Baretta wrote:
> > What are the differences between Unix.select and Thread.select? Is it
> > safe to use the former in a multithreaded program?
>
> Without looking at the implementation, I would guess that a
> Unix.select with a non-zero timeout would cause the whole process to
> block - whereas an equivalent call to Thread.select would only cause
> the calling thread to block.
Once upon a time, this was the intention of Thread.select. In current
ocaml, one can also call Unix.select, and only the calling thread is
blocked.
Archive: http://caml.inria.fr/pub/ml-archives/caml-list/2005/03/8427a70b9f6d461c56c18109657fa224.en.html
Oliver Bandel asked and Sachin Shah answered:
> I found the documentation on GraphPS via CamlHump,
> but the ftp-adress which was mentioned in the documentation
> was unvalid.
>
> Where to find GraphPS?
GraphPS 1.0 and 1.1 are available at:
ftp://ftp.inria.fr/INRIA/Projects/cristal/caml-light/bazar-ocaml/
Archive: http://caml.inria.fr/pub/ml-archives/caml-list/2005/03/87cab0cdbbcfcf736b85d746ba610e78.en.html
Jacques Carette announced:
Attached is a camlp4 syntax extension for a 'monadic do notation' for OCaml,
much like Haskell's. The syntax extension is joint work with Oleg Kiselyov,
and is loosely based on previous work of Lydia van Dijk on a similar syntax
extension.
Naturally, in OCaml certain monads (like State and IO) are unnecessary. But
other monads (List, option, etc) can be quite convenient.
But monads can be much more than convenient. Below is some code written in a
non-deterministic monad of streams. This example is particularly interesting
in that it cannot be (naively) done in either Prolog or in Haskell's MonadPlus
monad, both of which would go into an infinite loop on this example.
(* test non-determinism monad, the simplest possible implementation *)
type 'a stream = Nil | Cons of 'a * (unit -> 'a stream) | InC of (unit -> 'a stream)
let test_nondet () =
let mfail = fun () -> Nil in
let ret a = fun () -> Cons (a,mfail) in
(* actually, interleave: a fair disjunction with breadth-first search*)
let rec mplus a b = fun () -> match a () with
| Nil -> InC b
| InC a -> (match b () with
| Nil -> InC a
| InC b -> InC (mplus a b)
| Cons (b1,b2) -> Cons (b1, (mplus a b2)))
| Cons (a1,a2) -> Cons (a1,(mplus b a2)) in
(* a fair conjunction *)
let rec bind m f = fun () -> match m () with
| Nil -> mfail ()
| InC a -> InC (bind a f)
| Cons (a,b) -> mplus (f a) (bind b f) () in
let guard be = if be then ret () else mfail in
let rec run n m = if n = 0 then [] else
match m () with
| Nil -> []
| InC a -> run n a
| Cons (a,b) -> (a::run (n-1) b)
in
let rec numb () = InC (mplus (ret 0) (mdo { n <-- numb; ret (n+1) })) in
(* Don't try this in Prolog or in Haskell's MonadPlus! *)
let tst = mdo {
i <-- numb;
guard (i>0);
j <-- numb;
guard (j>0);
k <-- numb;
guard (k>0);
(* Just to illustrate the `let' form within mdo *)
let test x = x*x = j*j + k*k in;
guard (test i);
ret (i,j,k)
} in run 7 tst
;;
We ourselves have been experimenting with a combined state-passing,
continuation-passing monad, where the values returned and manipulated are code
fragments (in MetaOCaml). This allows for considerably simpler, typed code
combinators. Details of this work will be reported elsewhere.
The attached file (unavailable from the list archives):
(* name: pa_monad.ml
* synopsis: Haskell-like "do" for monads
* authors: Jacques Carette and Oleg Kiselyov,
* based in part of work of Lydia E. Van Dijk
* last revision: Sun Mar 27 2005
* ocaml version: 3.08.0 *)
(** Conversion Rules
Grammar informally:
mdo { exp }
mdo { exp1; exp2 }
mdo { x <-- exp; exp }
mdo { let x = foo in; exp }
which is almost literally the grammar of Haskell `do' notation,
modulo `do'/`mdo' and `<-'/`<--'.
Grammar formally:
mdo { <do-body> }
<do-body> :: =
"let" var = EXP ("and" var = EXP)* "in" ";" <do-body>
EXP
(pat <--)? EXP ";" <do-body>
Semantics (as re-writing into the core language)
mdo { exp } ===> exp
mdo { pat <-- exp; rest } ===> bind exp (fun pat -> mdo { rest })
mdo { exp; rest } ===> bind exp (fun _ -> mdo { rest })
mdo { let pat = exp in; rest } ===> let pat = exp in mdo { rest }
Actually, in `let pat = exp' one can use anything that is allowed
in a `let' expression, e.g., `let pat1 = exp1 and pat2 = exp2 ...'.
The reason we can't terminate the `let' expression with just a semi-colon
is because semi-colon can be a part of an expression that is bound to
the pattern.
It is possible to use `<-' instead of `<--'. In that case,
the similarity to the `do' notation of Haskell will be complete. However,
due to the parsing rules of Camlp4, we would have to accept `:=' as
an alias for `<-'. So, mdo { pat := exp1; exp2 } would be allowed too.
Perhaps that is too much.
The major difficulty with the `do' notation is that it can't truly be
parsed by an LR-grammar. Indeed, to figure out if we should start
parsing <do-body> as an expression or a pattern, we have to parse it
as a pattern and check the "<--" delimiter. If it isn't there, we should
_backtrack_ and parse it again as an expression. Furthermore, "a <-- b"
(or "a <- b") can also be parsed as an expression. However, for some patterns,
e.g. (`_ <-- exp'), that cannot be parsed as an expression.
*)
type monbind = BindL of (MLast.patt * MLast.expr) list
| BindM of MLast.patt * MLast.expr
| ExpM of MLast.expr
(* Convert MLast.expr into MLast.patt, if we `accidentally'
parsed a pattern as an expression.
The code is based on pattern_eq_expression in
/camlp4/etc/pa_fstream.ml *)
let rec exp_to_patt loc e =
match e with
<:expr< $lid:b$ >> -> <:patt< $lid:b$ >>
| <:expr< $uid:b$ >> -> <:patt< $uid:b$ >>
| <:expr< $e1$ $e2$ >> ->
let p1 = exp_to_patt loc e1 and p2 = exp_to_patt loc e2 in
<:patt< $p1$ $p2$ >>
| <:expr< ($list:el$) >> ->
let pl = List.map (exp_to_patt loc) el in
<:patt< ($list:pl$) >>
| _ -> failwith "This pattern isn't yet supported"
(* The main semantic function *)
let process loc b =
let globbind2 x p acc =
<:expr< bind $x$ (fun $p$ -> $acc$) >>
and globbind1 x acc =
<:expr< bind $x$ (fun _ -> $acc$) >>
and ret n = <:expr< $n$ >> in
let folder = let (a,b) = (globbind2, globbind1) in
(fun accumulator y ->
match y with
| BindM(p,x) -> a x p accumulator
| ExpM(x) -> b x accumulator
| BindL(l) -> <:expr< let $list:l$ in $accumulator$ >>
)
in
match List.rev b with
| [] -> failwith "somehow got an empty list from a LIST1!"
| (ExpM(n)::t) -> List.fold_left folder (ret n) t
| _ -> failwith "Does not end with an expression"
EXTEND
GLOBAL: Pcaml.expr;
Pcaml.expr: LEVEL "expr1"
[
[ "mdo"; "{";
bindings = LIST1 monadic_binding SEP ";"; "}" ->
process loc bindings
]
] ;
Pcaml.expr: BEFORE "apply"
[ NONA
[ e1 = SELF; "<--"; e2 = Pcaml.expr LEVEL "expr1" ->
<:expr< $e1$ $lid:"<--"$ $e2$ >>
]
] ;
monadic_binding:
[
[ "let"; l = LIST1 Pcaml.let_binding SEP "and"; "in" ->
BindL(l) ]
|
[ x = Pcaml.expr LEVEL "expr1" ->
(* For some patterns, "patt <-- exp" can parse
as an expression too. So we have to figure out which is which. *)
match x with
<:expr< $e1$ $lid:op$ $e3$ >> when op = "<--"
-> BindM((exp_to_patt loc e1),e3)
| _ -> ExpM(x) ]
|
[ p = Pcaml.patt LEVEL "simple"; "<--"; x = Pcaml.expr LEVEL "expr1" ->
BindM(p,x) ]
] ;
END;
Walid Taha said and Jacques Carette answered:
> This is pretty cool!
Thanks.
> Are you using the same monad we used in the FFT work?
Yes. For those on caml-list, the paper Walid refers to is
@inproceedings{KiselyovSwadiTaha2004,
author = {Oleg Kiselyov and Kedar N.~Swadi and Walid Taha},
title = {A methodology for generating verified combinatorial
circuits},
booktitle = {ACM International conference on Embedded Software
(EMSOFT)},
year = {2004}
}
> Also, can you either get rid of the ";" after the "in" or the "in" before
the ";"? :)
See the comments in pa_monad.ml. Unfortunately we cannot easily do that; in
fact the required amount of work to achieve this might be quite large! It
would be by far easier to integrate this right into the grammar rather than
to post-facto extend the grammar via camlp4. At least, as far as my current
understanding of camlp4 goes -- maybe if a lot more of the tricks of the
camlp4 Masters were documented, it would not look so hard anymore!
William Harrison said:
I've been reading your recent posts, and given the interest in stream (a.k.a.
resumption) monads, I thought you might be interested in a draft I've submitted
about the use of just such monads:
http://www.cs.missouri.edu/~harrison/drafts/CheapThreads.pdf
This article describes the construction of a multitasking operating system
kernel in Haskell 98 using resumption monads; the kernel has all the "usual" OS
behaviors (e.g., process forking, preemption, message passing, synchronization,
etc.) captured succintly.
Here is a quick trick to help you read this CWN if you are viewing it using vim (version 6 or greater).
:set foldmethod=expr
:set foldexpr=getline(v:lnum)=~'^=\\{78}$'?'<1':1
zM
If you know of a better way, please let me know.
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.