Re: external versions of sh built-in utilities

[Robert Elz <kre%munnari.OZ.AU@localhost>]

    Date:        Mon, 17 Jul 2017 15:35:38 +0200
    From:        Edgar =?iso-8859-1?B?RnXf?= <ef%math.uni-bonn.de@localhost>
    Message-ID:  <20170717133538.GW50210%trav.math.uni-bonn.de@localhost>

  | Are we talking about (in POSIX speak) "utilities implemented as a built-in"

No.

  | (as opposed to "built-in utilities" 

Those.

  | -- which, in turn, differ from "special built-in utilities")?

They do.

  | Examples of (regular) built-in utilities are true, cd.

Yes, though there is no rational reason (than an accident of ksh88 history)
for true or false to be in this category (there are issues waiting to be
adequately addressed related to how built-ins are found, etc, in which the
question of how true & false should be characterised is a side-issue, so
they may eventually get moved to where they belong, which is...)

  | Examples of utilities implemented as a built-in may be echo, printf.

yes.   (pwd ought be in this category as well, and kill is a weird 
intermediate case.)

And from the later message:

  | As this is in the "introduction" section and there's a more precise (and
  | more sensible) definition in 2.9.1, I would rather file an error with
  | the Austin Group than trying to implement the words of 1.6. 

The two are not in-consistent in any way, and 1.6 imposes a conformance
requirement.

Yes, it is (mostly) stupid (though there are some kind of weird things that
can be done using find, or xargs, that don't work unless things like "cd"
and read are exec'able (though of course they could be accomplished using
"sh -c 'read ..'" etc) - I can't find any rational excuse for wait/jobs/fg/...
however.

In any case the message I have taken from all the replies is that NetBSD
will not be implementing a funky script in order to satisfy this requirement,
and it will be just one more place where we choose non-compliance in the
name of common sense.

kre