Re: set -vx changes output of subshells

[Peter Stephenson <pws@ifh.de>]

Daniel Dignam wrote:
> Hi Folks,
> 
>   I've just built zsh 3.1.1 and have found a problem with redirection in
> subshells and set -vx. It appears to be adding a return at the start of
> the output (this makes debugging a trifle hard):
> 
> alfep3% set -vx
> alfep3% a=$(ls foo 2>&1)
> a=$(ls foo 2>&1)
> + a=+ a=+ ls foo
> foo 
> alfep3% echo $a | wc -l
> echo $a | wc -l
> + echo + a=+ ls foo
> foo
> + wc -l
>          2

I got a chance to think about this (didn't notice any other replies)
and it's not really clear to me it's a bug.  (Sorry, I know that
answer grates on the nerves.)  You've told zsh to send a message of
what it's doing to stderr, then you've redirected stderr from $(...)
to the output... naturally the xtrace output goes to the variable so
that a contains an initial line `+ a=+ ls -F -l zsh'.  

Ksh does something different: looking at the output, it seems to be
because the 2>&1 doesn't get processed till last, but I can't see a
good reason why the shell should *have* to do that.  It could probably
be moved further up in zsh, removing the doubled xtrace code for
builtins and external commands, though it looks like the addvars()
part would still have to be separate.  This would presumably fix this
problem.

There are certainly still minor problems with set -x, though: for
example, the 2>&1 doesn't appear at all.  Probably something needs to
be added to addfd().

I noticed one other thing with the option handling: the EXECOPT
option, properly the noexec option except it's negated, seems to be
doing rather to much work: it does all the forking and I/O redirection
even though it doesn't actually execute the commands.  So, for
example, a script containing:

print this is a file >shfile
set -n
print this is no longer a file >shfile

in ksh will leave shfile containing `this is a file', in zsh will
leave an empty shfile, since the redirection has been done for the
second print even though the command wasn't executed.

This is presumably wrong, but how little should noexec be doing?  Can
we omit the fork in execcmd()?  Should the glob then be done without a
fork even if the command not to be executed had a & after it?

-- 
Peter Stephenson <pws@ifh.de>       Tel: +49 33762 77366
WWW:  http://www.ifh.de/~pws/       Fax: +49 33762 77413
Deutsches Elektronen-Synchrotron --- Institut fuer Hochenergiephysik Zeuthen
DESY-IfH, 15735 Zeuthen, Germany.