On 3/18/10, Martin Jambon <martin.jambon@ens-lyon.org> wrote:
> Stas Miasnikou wrote:
>> OCaml 3.11.1, OpenBSD 4.6, i386.
>>
>> I am trying to read whole file by doing:
>>
>> let read_file_bin name =
>>   let ic = open_in_bin name in
>>   let b = Buffer.create 1024 in
>>   (try Buffer.add_channel b ic max_int with _ -> ()); (* <-- HERE *)
>>   close_in ic;
>>   Array.init (Buffer.length b) (fun i -> int_of_char (Buffer.nth b i))
>>
>> but it hangs on the line marked. Am I doing something wrong?
>
> The problem is max_int and the fact that Buffer.add_channel and
> Buffer.resize
>  do not check for this possibility:
>
> let add_channel b ic len =
>   if b.position + len > b.length then resize b len;
>   really_input ic b.buffer b.position len;
>   b.position <- b.position + len
>
> Something like the following would be better:
>
> let add_channel b ic len =
>   if len < 0 || len > Sys.max_string_length then
>     invalid_arg "Buffer.add_channel";
>   ...

Oh, never thought OCaml has bugs! ;-)

> Since you uncovered this problem, please kindly submit a proper bug report
> at
>   http://caml.inria.fr/mantis

I've submitted it.

> (and figure what to do if the file is larger than 16MB on 32-bit systems)
>
> Of course, you can see from the implementation of the Buffer module that a
> string of your maximum length is created no matter what, which you surely
> want
> to avoid especially on 64-bit systems where Sys.max_string_length is very
> large.

Erm... yes. I think I follow the other advice and will use Bigarray.

Stas