How can I use SYS:OPEN-PIPE with programs that read until EOF,
i.e. how can I signal that I'm finished sending data to a program? The
example from the docs doesn't work in this case:
(with-open-stream
(s (sys:open-pipe "perl -lne '$count++; END {print $count}'"
:direction :io))
(write-line "foo" s)
(write-line "bar" s)
(force-output s)
(read-line s))
I want this to return "2" but it hangs instead.
Thanks,
Edi.
On Tuesday 28 October 2003 06:26 am, Edi Weitz wrote:
> How can I use SYS:OPEN-PIPE with programs that read until EOF,
....
> (with-open-stream
> (s (sys:open-pipe "perl -lne '$count++; END {print $count}'"
....
> I want this to return "2" but it hangs instead.
If you run the line of perl by itself in bash, you find that it does not
finish until you hit ^D (unix eof character).
The lisp code and the perl code are hanging, waiting for each other to send
EOF.
This is a protocol problem.
You either need to open two separate streams - one output, one input - and
close the output pipe before reading from the input, or, you need to rewrite
the perl scriptlet so that it runs a simple protocol and "knows" when to quit
(e.g. it quits when lisp sends it some specific character).
pt
p.s. After a quick glance at the manuals, I don't see how to open two streams
on one shell command, so I would suggest altering the perl.
On Tue, 28 Oct 2003 11:08:40 -0500, tarvydas <tarvydas@allstream.net> wrote:
> You either need to open two separate streams - one output, one input
> - and close the output pipe before reading from the input, or, you
> need to rewrite the perl scriptlet so that it runs a simple protocol
> and "knows" when to quit (e.g. it quits when lisp sends it some
> specific character).
>
> p.s. After a quick glance at the manuals, I don't see how to open
> two streams on one shell command, so I would suggest altering the
> perl.
Altering the program is not an option, unfortunately. (It's a C
program and I don't have the source. I used the Perl snippet for
illustration purposes only.)
I had hopes that LW would allow for separate streams like CMUCL or
AllegroCL but it looks like this isn't possible. In CMUCL you can, for
example, do it like this:
(let (prog)
(unwind-protect
(progn
(setq prog
(ext:run-program "/tmp/foo.pl" nil
:output :stream
:input :stream
:wait nil))
(with-open-stream (in (ext:process-input prog))
(write-line "foo" in)
(write-line "bar" in)
(write-line "baz" in))
(with-open-stream (out (ext:process-output prog))
(with-output-to-string (s)
(loop for line = (read-line out nil nil)
while line
do (write-string line s)))))
(ext:process-close prog)))
("foo.pl" is a small Perl script which counts the lines coming from
stdin. This form will actually return the string "3".)
This doesn't work in CMUCL, BTW, if I replace "foo.pl" with the Perl
one-liner from my previous post but this doesn't bother me. I guess
you have to use the :PTY variant of RUN-PROGRAM in this case.
Thanks,
Edi.
On Tue, 28 Oct 2003 17:12:13 -0500, tarvydas <tarvydas@allstream.net> wrote: > Pending a reply from David Fox, have you tried the following kludges?: > > a) if you only need the final reply from the C program, then use an > input stream and redirect the output of the C program to a file > (sys:with-open-pipe "cprog >/tmp/reply" :direction :output) > > b) Write a small C / Perl / whatever program that invokes the big C > program, redirecting stdin and stdout to two separate file > descriptors. The tiny program can run a small protocol with the > lisp program. > > c) You could probably do it all with FFI and direct calls to Unix > exec, read and write. > > d) Use two comm:socket-stream's in lisp. Write two perl (C / Tcl/Tk > / whatever) helpers - one which reads from a socket and writes to > stdout, the other reads from stdin and writes to a socket, then do > something like > > (sys:run-shell-command "sock2stdout | cprog | stdin2sock") Thanks for the helpful ideas! I'll try some of these unless someone comes up with a LW feature I don't know of yet. Thanks again, Edi.