Re: "typeset -p" inconsistency

[Peter Stephenson <p.w.stephenson@xxxxxxxxxxxx>]

> On 01/11/2022 05:00 Bart Schaefer <schaefer@xxxxxxxxxxxxxxxx> wrote:
> % setopt extendedglob
> % typeset -U CDPATH
> % typeset -mp '(#i)cdpath'
> typeset -aT CDPATH cdpath=(  )
> typeset -UT CDPATH cdpath=(  )
> 
> Note the -U attribute is only retained when printing the assignment
> for the scalar ... which is literally correct, since it was only
> asserted for the scalar, but lossy, whereas the array-ness is
> preserved in both cases (as long as typeset remains a reserved word).

That does correctly reflect the behaviour of the -U flag, however.

% typeset -T FOO foo
% foo=(one two three)   
% typeset -U foo
% print $foo
one two three
% foo+=(two)           # flag applied
% print $foo
one two three
% FOO+=:two            # flag not applied
% print $foo
one two three two

Other consequences of this are logical, e.g. "foo=($foo)" does
prune, "typeset -U FOO" works as expected.

The ideal output for typeset -p to reflect how this works might
be to output the typeset -T separately from any indication
of any flags set on one or other of the two interfaces to the
value, i.e. something like what I typed (but with the current
value somewhere).

typeset -T FOO foo=(one two three)
typeset -U foo

pws