Re: Insidious exit status bugs

[Peter Stephenson <pws@xxxxxxxxxxxxxxxxx>]

"Bart Schaefer" wrote:
> The following really messed me up, trying to write a script that will work
> whether sh is bash or zsh.  Here's zsh 3.0.5 (3.1.4 is the same):
> 
> zagzig% echo yyy | fgrep -q xxx && echo ok
> zagzig% echo yyy | fgrep -q `echo xxx` && echo ok
> ok
> 
> It appears that the exit status of `echo xxx` is masking the exit status of
> `fgrep -q`, but I'm not certain.

Insidious is definitely right.  This is serious harrow-up-thy-soul
territory.  There appear to be two things happening.  The first I
think I understand, though not why it's there.  The second currently
has me stumped.

First: the exit status from the `...` or $(...) is being explicitly
passed back as lastval, i.e. $?.  Why???  There's a variable cmdoutval
whose only function in life is to remember that status to be assigned
to lastval, so someone thought it was a good idea.  Here's the
offending chunk from getoutput() in exec.c which is for the parent
side of the forked $(...).

    } else if (pid) {
	LinkList retval;

	zclose(pipes[1]);
	retval = readoutput(pipes[0], qt);
	fdtable[pipes[0]] = 0;
	child_suspend(0);		/* unblocks */
	lastval = cmdoutval;
	return retval;
    }

`lastval = cmdoutval' is deliberately messing up the return status for
reasons I can't even guess at.  Perhaps if the second bug wasn't
there, this wouldn't be seen, but it had me confused anyway.


Second:  I commented the cmdoutval stuff out, and now things start to
go really haywire.  It looks like you have to have a builtin in the
first part of the pipe, and an external command in the second part to
see this.  Bearing this in mind, I commented out the chunk in
execpline2() where `builtin as first command in pipe' is specially
handled, so as to simplify things, and added some messages; it still
goes wrong:

+11:58% echo yyy | fgrep -q $(echo xxx)
Got status 0 from process 36923               OK, that's the $(echo xxx)
...was procsub process.
Got status 0 from process 36410               This must be the echo yyy...
Setting status of 36410 to 0 (1)
Exit status 0 from 36410                      ...and this means that
                                                 it's being handled
                                                 as if it were last in
                                                 the pipeline.
setting lastval to 0 (1)
Got status 0 from process -1
...error: No child processes                  self explanatory, but...
+11:58% Got status 256 from process 36412     now we're back to the
                                              prompt, and 36412 (fgrep)
                                              has only just been handled.
Got status 256 from process -1
...error: No child processes


So it looks like the fgrep is being run, or at least waited for,
asynchronously instead of synchronously, and all because of the $(echo
xxx).  I've wasted too much time on this already, so that's it for
now.

-- 
Peter Stephenson <pws@xxxxxxxxxxxxxxxxx>       Tel: +39 050 844536
WWW:  http://www.ifh.de/~pws/
Gruppo Teorico, Dipartimento di Fisica
Piazza Torricelli 2, 56100 Pisa, Italy