Re: fileexist(f)

[Ethan M. <merritt@u.washington.edu>]

On Wednesday 08 March 2006 01:26 pm, Petr Mikulik wrote:
> >>    pipeexist(s) which returns 1 for "-" and 1/0 whether it is a
> >> pipe wherefrom you can read at least 1 byte;
> >
> > Huh?  But that is quite broken.  If you read from that data source,
> > even 1 byte, then you have poisoned it for actual plotting.
>
> I don't think so. plot '<awk {blalbal} bla.dat' gives always the
> same. Can you give an counter-example?

Any real-time data source?
A keyboard input stream?
A pipe to another interactive program?

> >> The truth is that popen() "never" returns NULL
> >> (contrary to specification) -- at least for OS/2, Windows, Linux.
> >
> > On linux it returns NULL if the pipe open fails.
>
> No, it does NOT! Try it!
> popen("bla bla", "r") does not return NULL

It does not return NULL in that case because it successfully opened
a pipe to the forked command "/bin/sh bla bla".   The forked shell
itself returned an error ("command not found").
But there is no pipe failure here.  

> I would prefer no system error message at all.

Then you must configure your shell environment to not
print an error.  This is being printed by /bin/sh, not by gnuplot.
The fact that it is printed at all is an indication that the popen()
did in fact succeed.

> But popen() did not return NULL. So I must read at least one byte.

You are looking in the wrong place for the error return.
If you want to do error checking, you need something like this:

--- gnuplot/src/datafile.c      2006-01-27 08:36:48.000000000 -0800
+++ gnuplot-cvs/src/datafile.c  2006-03-08 14:25:53.000000000 -0800
@@ -1343,7 +1343,9 @@
     if (!mixed_data_fp) {
 #if defined(PIPES)
         if (df_pipe_open) {
-            (void) pclose(data_fp);
+            int ierr = pclose(data_fp);
+           if (ierr)
+               fprintf(stderr,"pipe error %s\n",strerror(ierr));
             df_pipe_open = FALSE;
         } else
 #endif /* PIPES */

With that patch in place, your test example gives:

gnuplot> plot "< bla bla"
  sh: bla: command not found
  pipe error Unknown error 32512

I'm not sure how to make it print a more interpretable error message
string, but I assume that 32512 in fact means "command not found".

> > Anyhow, my main point is that an   xxx_exists()  routine should be
> > exactly parallel to the routine used for actual data input. 
> > Otherwise you cannot trust the result as an indication of whether
> > you will or will not be able to plot it.   Thus it should handle
> > both files and pipes, because the data input routine handles both
> > files and pipes.
>
> Then I propose   streamexist()  that would do all of that; it would
> be equivalent to
>  	if (fileexist(file_in_path(x)) || pipeexist(x))

OK, but the pipe part of the test must not actually read any data.


-- 
Ethan A Merritt
Biomolecular Structure Center
University of Washington, Seattle WA