Re: Elusive pipe I/O bug in Scsh?

[Stefan Jankowski <dave.null@spambucket.informatik.uni-freiburg.de>]

>>>>> "Michael" == Michael Sperber [Mr Preprocessor] 
>>>>> <sperber@informatik.uni-tuebingen.de> writes:

    Michael> The trivial question is whether you're sure that your
    Michael> helper program always flushes its output to point where a
    Michael> complete S-expression results.

Is this a problem? Wouldn't (read) block anyway until it has read a
complete sexp? The following at least seems to work:

,----
| lunaris:~/sunray/ufr/scsh% cat read2.scm
| #!/usr/local/bin/scsh -s
| !#
| 
| (write (read))
| (newline)
| 
| (write (read))
| (newline)
| 
| lunaris:~/sunray/ufr/scsh% cat write2.c
| 
| #include <stdio.h>
| 
| int main(void)
| {
|     puts("("); fflush(stdout);
|     puts("foo"); fflush(stdout);
|     puts("."); fflush(stdout);
|     fputs("ba", stdout); fflush(stdout);
|     puts("r"); fflush(stdout);
|     puts(") (Foo ."); fflush(stdout);
|     fputs("Ba", stdout); fflush(stdout);
|     puts("r"); fflush(stdout);
|     puts(")");
| 
|     return 0;
| }
| lunaris:~/sunray/ufr/scsh% gcc -Wall write2.c -o write2
| lunaris:~/sunray/ufr/scsh% ./write2 | ./read2.scm
| (foo . bar)
| (Foo . Bar)
| lunaris:~/sunray/ufr/scsh% 
`----

Flushing within sexps/tokens or omitting flushing between sexps does
not seem to do any harm, and this is what I would have expected. Or do
you mean that scsh fails to (read) sexps properly if the writer
blocks because the pipe's buffer runs full within a sexp? That doesn't
seem to be a problem either, and I don't think it should be:

,----
| lunaris:~/sunray/ufr/scsh% cat read3.scm
| #!/usr/local/bin/scsh \
| -dm -o read3 -e start -s
| !#
| 
| (define-structure read3
| 
|   (export start)
| 
|   (open scheme scsh threads)
| 
|   (begin
| 
|     (define (start args)
|       (sleep 5000)
|       (write (length (read)))
|       (newline))))
| lunaris:~/sunray/ufr/scsh% cat write3.c
| 
| #include <stdio.h>
| 
| int main(void)
| {
|     int i;
| 
|     fputs("(", stdout);
|     for (i = 0; i < 12000; i++)
|       fputs("foo ", stdout);
|     puts(")");
| 
|     return 0;
| }
| lunaris:~/sunray/ufr/scsh% gcc -Wall write3.c -o write3
| lunaris:~/sunray/ufr/scsh% ./write3 | ./read3.scm
| 12000
| lunaris:~/sunray/ufr/scsh% 
`----

My understanding is that (read) treats its input as a simple stream of
octets without regard to linebreaks (other than treating them as
whitespace) or blocking in OS-level reads; or am I missing something?

    Michael> Other than that, your best bet is probably posting as
    Michael> much of the code as possible so we can try to reproduce
    Michael> the problem.

I'll see what I can do.

Oh yes, the Scsh version is 0.6.3

Regards,
Stefan

-- 
Stefan Jankowski
University of Freiburg, CS Dpt.
s/spambucket//
s/dave\.null/janky/