clorb-devel Mailing List for Common Lisp ORB

>>>>> "ls" == Lennart Staflin <le...@ly...> writes:

  ecm> ok, I think the problem is in the CMUCL definition of
  ecm> read-sequence on socket streams. After replacing it by a simple
  ecm> version, everything works as expected.

  ls> Does it depend on the CMUCL version. The CMU Common Lisp release
  ls> x86-linux 2.4.19 8 February 2000 build 455, works fine for me on
  ls> Linux x86.

I don't know; I only have access to CMUCL on Solaris. I reported the
problem to the CMUCL maintainers, who have confirmed that it is a bug.
You may not see it on Linux: the problem seems to be that
#'read-sequence is mapped to a read(), so if the platform read() is
able to keep up with the socket you won't see the problem.

-- 
Eric Marsden                          <URL:" rel="nofollow">http://www.laas.fr/~emarsden>

Eric Marsden <ema...@ma...> writes:

> >>>>> "ecm" == Eric Marsden <ema...@ma...> writes:
> 
>   ecm> I get the same error whether using native CMUCL sockets, or my
>   ecm> Solaris port of db-sockets.
> 
> ok, I think the problem is in the CMUCL definition of read-sequence on
> socket streams. After replacing it by a simple version, everything
> works as expected.

Does it depend on the CMUCL version. The CMU Common Lisp release
x86-linux 2.4.19 8 February 2000 build 455, works fine for me on Linux
x86.

-- 
Lennart Staflin  <le...@ly...>

>>>>> "ecm" == Eric Marsden <ema...@ma...> writes:

  ecm> I get the same error whether using native CMUCL sockets, or my
  ecm> Solaris port of db-sockets.

ok, I think the problem is in the CMUCL definition of read-sequence on
socket streams. After replacing it by a simple version, everything
works as expected.

   (defun my-read-sequence (sequence stream &key (start 0) (end nil))
     (unless end (setq end (length sequence)))
     (do ((index start (1+ index)))
         ((eql index end) index)
       (let ((b (read-byte stream nil :eof)))
         (when (eq b :eof) (return index))
         (setf (aref sequence index) b))))
 
There are two calls to read-sequence in clorb-request.lisp to change.  
  
-- 
Eric Marsden                          <URL:" rel="nofollow">http://www.laas.fr/~emarsden>

>>>>> "bc" == Brad Chapman <cha...@ar...> writes:

  bc> ; Loading #p"/usr/home/chapmanb/clorb-0.2/test-intern.lisp".
  bc> Warning: This variable is undefined:
  bc>   *R
  bc> Warning: This function is undefined:
  bc>   INTERNALIZE

you need to (load "internalize") before running that test.

  
  bc> Could the problem be due to sockets? (since that seems to be a
  bc> sticking point with the clisp server). The error didn't really
  bc> seem like it, but I am also still working on getting Dan
  bc> Barlow's socket code working on FreeBSD, so I am using whatever
  bc> socket capabilities cmu cl has. Just a random idea because I'm
  bc> not positive where to go. To talk myself out of that, if Eric
  bc> has this problem as well, this may not be the case.

I get the same error whether using native CMUCL sockets, or my Solaris
port of db-sockets. 
  

  bc> *** - UNIX error 49 (EADDRNOTAVAIL): Can't assign requested 
  bc> address

as Lennart said, my idea of the socket already being in use was
probably wrong. The error could come from trying to bind to a socket
on an address which doesn't belong to the local machine. I don't think
the error is in CLORB, since it just does (socket-server port); maybe
it's a problem with CLISP. You might be able to find out what's going
on using strace.

-- 
Eric Marsden                          <URL:" rel="nofollow">http://www.laas.fr/~emarsden>