Bug 1448950

Description of problem:

The fix for 1072086 introduces at least two bugs (that I know of).
Maybe more.

Regression #1:
========== ==

If you have "set -x" turned on, and then redirect the stderr of some
command into a file, the output file to which only the command's stderr
should have been redirected now contains the trace data that should have
gone to the script's stderr instead of the comand's stderr.  Here, it's
easier to show than describe:

0 vm38(ROOT)/root ### rpm -q mksh
mksh-39-9.el6.x86_64
0 vm38(ROOT)/root ### cat /tmp/dummy
#!/bin/mksh

set -x

echo Yadda. > /tmp/yadda 2>&1
0 vm38(ROOT)/root ### /tmp/dummy ; echo ======================== ; cat /tmp/yadda
+ echo Yadda.
+ > /tmp/yadda
+ 2>&1
========================
Yadda.
0 vm38(ROOT)/root ###

Now, that's working more or less how I'd expect with the previous
version of mksh.  Here's what happens with the new version:

0 vm38(ROOT)/root ### rpm -q mksh
mksh-39-10.el6.x86_64
0 vm38(ROOT)/root ### /tmp/dummy ; echo ======================== ; cat /tmp/yadda
+ > /tmp/yadda
+ 2>&1
========================
+ echo Yadda.
Yadda.
0 vm38(ROOT)/root ###

Note the "+ echo Yadda." ended up in the command's stderr instead of the
script's stderr.

Regression #2:
========== ==

I think this is probably the real problem, and the above is just an
alternate manifestation of the same thing.  When I take the redirection
of stderr out of the script:

0 vm38(ROOT)/root ### rpm -q mksh
mksh-39-10.el6.x86_64
0 vm38(ROOT)/root ### cat /tmp/dummy
#!/bin/mksh

set -x

echo Yadda. > /tmp/yadda
0 vm38(ROOT)/root ### /tmp/dummy ; echo ======================== ; cat /tmp/yadda
+ > /tmp/yadda
+ echo Yadda.
========================
Yadda.
0 vm38(ROOT)/root ###

Note the order of the two trace lines.  Here's what it used to do:

0 vm38(ROOT)/root ### rpm -q mksh
mksh-39-9.el6.x86_64
0 vm38(ROOT)/root ### /tmp/dummy ; echo ======================== ; cat /tmp/yadda
+ echo Yadda.
+ > /tmp/yadda
========================
Yadda.
0 vm38(ROOT)/root ###

The command used to come out before the redirection, now it's the
other way around.  If I'd have it my way, they'd all be on one line,
but meh, the wole world doesn't cater to me.  Anyway, I can see that
causing major confusion.  I suspect, though, that the two problems
are related.  I'm thinking that by getting them out of order like that,
we managed to also generate the trace output after setting up stdout
and stderr redirection for commands.

Version-Release number of selected component (if applicable):

mksh-39-10.el6

How reproducible:

Every time.

Steps to Reproduce:
(see above)

Actual results:
(see above)


Expected results:
(see above)