Named pipe problem: is this a bug?

Named pipe problem: is this a bug?

[Alan Schmitt]

[-- Attachment #1: Type: text/plain, Size: 894 bytes --]

Hello,

I've stumbled upon a small bug, and I wonder if it's a OS X bug (I'm  
using X.4.2), or a caml bug. The problem is the following: I create a  
named pipe, then I open it. The open system call seems to block until  
there is data in the pipe.

Here is a small program illustrating this:

let _ =
   Unix.mkfifo "pipe_test" 0o640;
   print_endline "pipe created";
   flush stdout;
   let chin = open_in "pipe_test" in
     print_endline "pipe opened";
     flush stdout

Compile it using "ocamlc unix.cma bug.ml" (the bug occurs both in  
bytecode and native code). When a.out is run, one sees:
pipe created
and the program keeps running. If one then does a
echo foo >> pipe_test
then the "pipe opened" line is shown, and the program terminates.

Is this behaviour normal?

Thanks,

Alan Schmitt

-- 
The hacker: someone who figured this out and made something cool happen.
.O.
..O
OOO



[-- Attachment #2: This is a digitally signed message part --]
[-- Type: application/pgp-signature, Size: 186 bytes --]

Re: [Caml-list] Named pipe problem: is this a bug?

[Virgile Prevosto]

Le 08/01/2005, à 05:03:42 PM, Alan Schmitt a écrit:
> let _ =
>    Unix.mkfifo "pipe_test" 0o640;
>    print_endline "pipe created";
>    flush stdout;
>    let chin = open_in "pipe_test" in
>      print_endline "pipe opened";
>      flush stdout
> 
> Compile it using "ocamlc unix.cma bug.ml" (the bug occurs both in  
> bytecode and native code). When a.out is run, one sees:
> pipe created
> and the program keeps running. If one then does a
> echo foo >> pipe_test
> then the "pipe opened" line is shown, and the program terminates.
> 
> Is this behaviour normal?
> 

Hello Allan,
this corresponds to what the manual page of fifo says (at least under
linux)
  The FIFO must be opened on both ends (reading and writing) before data
  can  be  passed. Normally, opening the FIFO blocks until the other end
  is opened also.

If you don't want that behavior, you'll have to open pipe_test in
non-blocking mode (e.g. with open_in_gen and the Open_nonblock flag)

By the way, what is the purpose of the Open_rdonly and Open_wronly
flags, given that the open_in* and open_out* functions return values of
different types?
-- 
E tutto per oggi, a la prossima volta
Virgile

Re: [Caml-list] Named pipe problem: is this a bug?

[Damien Bobillot]

[-- Attachment #1: Type: text/plain, Size: 1513 bytes --]


Le 1 août 05 à 17:03, Alan Schmitt a écrit :

> Hello,
>
> I've stumbled upon a small bug, and I wonder if it's a OS X bug  
> (I'm using X.4.2), or a caml bug. The problem is the following: I  
> create a named pipe, then I open it. The open system call seems to  
> block until there is data in the pipe.
>
> Here is a small program illustrating this:
>
> let _ =
>   Unix.mkfifo "pipe_test" 0o640;
>   print_endline "pipe created";
>   flush stdout;
>   let chin = open_in "pipe_test" in
>     print_endline "pipe opened";
>     flush stdout
>
> Compile it using "ocamlc unix.cma bug.ml" (the bug occurs both in  
> bytecode and native code). When a.out is run, one sees:
> pipe created
> and the program keeps running. If one then does a
> echo foo >> pipe_test
> then the "pipe opened" line is shown, and the program terminates.

Same thing on my computers (Mac OS X & Linux).

In fact, a pipe is considered to be opened only if both ends exists  
(you've created the read end, "echo foo >> .." create the write end).

http://www.annodex.net/cgi-bin/man/man2html?fifo+4 says :
"The kernel maintains exactly one pipe object for each FIFO special  
file that is opened by at least one process. The FIFO must be opened  
on both ends (reading and writing) before data can be passed.  
Normally, opening the FIFO blocks until the other end is opened also."

(it's no use to call flush after print_endline : printf_endline do  
the flush itself)

-- 
Damien Bobillot


[-- Attachment #2: smime.p7s --]
[-- Type: application/pkcs7-signature, Size: 2375 bytes --]

Re: [Caml-list] Named pipe problem: is this a bug?

[Richard Jones]

This is documented behaviour.  You can't open a FIFO in read mode
until a writer has opened the FIFO - instead the open(2) syscall will
block.  You can pass the O_NONBLOCK flag to open, in which case the
open call will not block.  Or you can open for read and write (because
you yourself are then supplying a writer).

Some more details here:

http://nixdoc.net/man-pages/Linux/man4/fifo.4.html

Rich.

-- 
Richard Jones, CTO Merjis Ltd.
Merjis - web marketing and technology - http://merjis.com
Team Notepad - intranets and extranets for business - http://team-notepad.com

Re: [Caml-list] Named pipe problem: is this a bug?

[Alan Schmitt]

[-- Attachment #1: Type: text/plain, Size: 254 bytes --]

Thanks for all the answers. I tried to man and apropos fifo, but I  
only have documentation about mkfifo which does not mention this issue.

Thanks again,

Alan

-- 
The hacker: someone who figured this out and made something cool happen.
.O.
..O
OOO



[-- Attachment #2: This is a digitally signed message part --]
[-- Type: application/pgp-signature, Size: 186 bytes --]