mksh R50d: parser fixes break common “set -u” workaround, take two

[Thorsten Glaser]

tglase@tglase:~ $ cat x
list_parts() {
        local regex=${2:-}
        local args=x
        if [ -n "$regex" ]; then
            args="${args} -regex '$regex'"
        fi
        printf '<%s>\n' $args
}
list_parts foo
tglase@tglase:~ $ bash x
<x>
tglase@tglase:~ $ env -i PATH=$PATH mksh x
typeset -i -U BASHPID
typeset -i COLUMNS
typeset EPOCHREALTIME
typeset -x HOME
typeset IFS
typeset -i -U KSHEGID
typeset -i -U KSHGID
typeset -i -U KSHUID
typeset -r KSH_VERSION
typeset -i LINES
typeset -i OPTIND
typeset -x PATH
typeset -i -U PGRP
set -A PIPESTATUS
typeset -i -r -U PIPESTATUS[0]
typeset -i -U PPID
typeset PS1
typeset PS2
typeset PS3
typeset PS4
typeset PWD
typeset -i -U RANDOM
typeset -i SECONDS
typeset -x SHELL
typeset -i TMOUT
typeset -i -U USER_ID
<x>

Breaks: php5 maintainer scripts

[Thorsten Glaser (mirabilos)]

Bad commit is:

commit 0cf0658a0d250604f896b98509718863a536f6d1
Author: tg <email address hidden>
Date: Tue Oct 7 15:22:17 2014 +0000

    fix severe regression in field splitting (LP#1378208)

*sigh*

[Thorsten Glaser (mirabilos)]

appears to be fixed by this patch. Everyone, *please test*!

[Thorsten Glaser (mirabilos)]

Testcase for this patch is in the description of this bug (some people on IRC wondered.)

I’ve not yet committed it. I’ll need to work on this more. But slowly, the “mystery” unravels.

----

Two more things to consider:

• We have several “word = quote || (!*x.str && (f & DOASNFIELD)) ? IFS_WORD : IFS_WS;” in the tree.
  Maybe changing them all to “[…] : IFS_IWS;” as this patch does is sensible?

• Maybe (cf. LP#1381993) DOASNFIELD should be “non-list context” instead?

[Thorsten Glaser (mirabilos)]

Preliminary fix (as posted) committed and uploaded to Debian.
Better fix committed, needs more testing.

[Thorsten Glaser (mirabilos)]

mksh R50e is out