caml-list - the Caml user's mailing list
 help / color / mirror / Atom feed
From: Gabriel Scherer <gabriel.scherer@gmail.com>
To: Martin Jambon <martin.jambon@ens-lyon.org>
Cc: OCaml Mailing List <caml-list@inria.fr>
Subject: Re: [Caml-list] How to mutate immutable record fields?
Date: Sat, 15 Oct 2011 11:15:44 +0200	[thread overview]
Message-ID: <CAPFanBGX3NAprKaqzxgsYpQmPiAU_1xyr_zMZohUYaH9AoKLiQ@mail.gmail.com> (raw)
In-Reply-To: <4E993FE5.7080409@ens-lyon.org>

I reached the same conclusions as Nicolas by looking at the lambda and
cmm output. I report my findings below, it case it could interest
other members of the list.

Here is the relevant part of the code provided by martin, hereafter 'test1':

 let t = create () in
 let f = Some 17 in
 Obj.set_field (Obj.repr t) 0 (Obj.repr f);

 let f1 = t.foo in
 let f2 = Obj.obj (Obj.field (Obj.repr t) 0) in

 if f1 = f then
   print_endline "OK     t.foo"
 else ...

Here is how you can "fix" it to make the test pass, hereafter 'test2':

 let t = create () in
 let f = Some 17 in
 Obj.set_field (Obj.repr t) 0 (Obj.repr f);

 let f1 = (fun t -> t.foo) t in
 let f2 = Obj.obj (Obj.field (Obj.repr t) 0) in

 if f1 = f then
   print_endline "OK     t.foo"
 else ...

Test1 and test2 -dlambda output both look sane:

  (let (t/1036 (apply (field 0 (global Test2!)) 0a) f/1037 [0: 17])
    (seq (array.unsafe_set (id t/1036) 0 (id f/1037))
      (let
        (f1/1038 (field 0 t/1036)   <= FIELD ACCESS
         f2/1039 (id (array.unsafe_get (id t/1036) 0)))
        (seq
          (if (caml_equal f1/1038 f/1037)  <= COMPARISON HERE
            (apply (field 29 (global Pervasives!)) "OK     t.foo")
            ....)))))
            (if (== f1/1038 0a)
              (apply (field 29 (global Pervasives!))
                "ERROR  t.foo is None")
              (apply (field 29 (global Pervasives!))
                "ERROR  t.foo is something strange")))

  (let (t/1036 (apply (field 0 (global Test!)) 0a) f/1037 [0: 17])
    (seq (array.unsafe_set (id t/1036) 0 (id f/1037))
      (let
        (f1/1038 (apply (function t/1039 (field 0 t/1039)) t/1036)
                  ^= FUNCTION CALL HERE
         f2/1040 (id (array.unsafe_get (id t/1036) 0)))
        (seq
          (if (caml_equal f1/1038 f/1037) <= SAME COMPARISON HERE
            (apply (field 29 (global Pervasives!)) "OK     t.foo")
            ...)))))

But the -dcmm change

  (let
    (t/1036 (app "camlTest2__create_1035" 1a addr)
     f/1037 "camlTest2__7")
    (if (!= (load unsigned int8 (+a t/1036 -4)) 254)
      (extcall "caml_modify" t/1036 f/1037 unit)
      (store float64u t/1036 (load float64u f/1037)))
    (let
        <= NOTHING HERE
      f2/1039
        (if (!= (load unsigned int8 (+a t/1036 -4)) 254) (load t/1036)
          (alloc 2301 (load float64u t/1036)))
      (if (!= (extcall "caml_equal" 1a f/1037 addr) 1)
                                    ^= 1a "CONSTANT" OPTIMIZED
        (app "camlPervasives__print_endline_1274" "camlTest2__6" unit)
        ...)))

  (let
    (t/1036 (app "camlTest__create_1035" 1a addr) f/1037 "camlTest__8")
    (if (!= (load unsigned int8 (+a t/1036 -4)) 254)
      (extcall "caml_modify" t/1036 f/1037 unit)
      (store float64u t/1036 (load float64u f/1037)))
    (let
      (f1/1038 (seq "camlTest__7" [] (load t/1036)) <= FUNCTION CALL HERE
       f2/1040
         (if (!= (load unsigned int8 (+a t/1036 -4)) 254) (load t/1036)
           (alloc 2301 (load float64u t/1036))))
      (if (!= (extcall "caml_equal" f1/1038 f/1037 addr) 1) <= COMPARISON
        (app "camlPervasives__print_endline_1274" "camlTest__6" unit)
        ....)))


On Sat, Oct 15, 2011 at 10:10 AM, Martin Jambon
<martin.jambon@ens-lyon.org> wrote:
> Dear fellow OCamlers,
>
>
> Hints on how to solve this, using any means necessary, will be greatly
> appreciated.
>
>
> Purpose: deserializing a record efficiently, i.e. creating a record
> whose field values are available one after another in an unpredictable
> order.
>
> Problem: Obj.set_field does the job in most cases but not in the example
> below. How to make it work?
>
> (we don't want to use pure OCaml option refs to store field values
> before putting them into the record because that's too slow)
>
> Requirements:
> - performance is critical
> - code will be machine-generated
> - immutable record fields must be supported
>
>
> The example below shows a record type for which a straight Obj.set_field
> produces a field that cannot be read correctly with the usual dot
> notation (ocamlopt only, tested on Linux/amd64 with OCaml 3.12.1 and
> 3.11.2).
>
>
> $ ocamlopt -o foo.opt foo.ml; ./foo.opt
> ERROR  t.foo is None
> OK     field0
>
> $ cat foo.ml
> (* ocamlopt -o foo.opt foo.ml; ./foo.opt *)
>
> type t = {
>  foo : int option;
>  bar : int;
> }
>
> let create () =
>  { { foo = None; bar = 42 } with foo = None }
>
> let () =
>  assert (create () != create ());
>  let t = create () in
>  let f = Some 17 in
>  Obj.set_field (Obj.repr t) 0 (Obj.repr f);
>
>  let f1 = t.foo in
>  let f2 = Obj.obj (Obj.field (Obj.repr t) 0) in
>
>  if f1 = f then
>    print_endline "OK     t.foo"
>  else if f1 = None then
>    print_endline "ERROR  t.foo is None"
>  else
>    print_endline "ERROR  t.foo is something strange";
>
>  if f2 = f then
>    print_endline "OK     field0"
>  else if f2 = None then
>    print_endline "ERROR  field0 is None"
>  else
>    print_endline "ERROR  field0 is something strange"
>
> --
> Caml-list mailing list.  Subscription management and archives:
> https://sympa-roc.inria.fr/wws/info/caml-list
> Beginner's list: http://groups.yahoo.com/group/ocaml_beginners
> Bug reports: http://caml.inria.fr/bin/caml-bugs
>
>


  parent reply	other threads:[~2011-10-15  9:16 UTC|newest]

Thread overview: 5+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2011-10-15  8:10 Martin Jambon
2011-10-15  9:07 ` Nicolas Pouillard
2011-10-15  9:13 ` Jeremy Yallop
2011-10-15  9:15 ` Gabriel Scherer [this message]
2011-10-18 21:45   ` Martin Jambon

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=CAPFanBGX3NAprKaqzxgsYpQmPiAU_1xyr_zMZohUYaH9AoKLiQ@mail.gmail.com \
    --to=gabriel.scherer@gmail.com \
    --cc=caml-list@inria.fr \
    --cc=martin.jambon@ens-lyon.org \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).