When should /bin/sh process ${ENV}

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

Bernd Ernesti <netbsd%lists.veego.de@localhost> said:
  |  It would be nice to have a summary on tech-userlevel%NetBSD.org@localhost 
and not in
  |  a pr.

The subject was PR [standards/42828] and anyone who wants the full discussion,
so far, and as much as it is, can find it all in that PR (which was filed
by Richard Hansen <rhansen%bbn.com@localhost>)

The substance of the PR was that NetBSD's /bin/sh runs ${ENV} for
all shells that start, whereas POSIX demands that it be run only
for interactive shells.

This isn't a bug in /bin/sh - what it is doing is exactly what
sh(1) says it will do.

The question is whether sh should be changed to comply with POSIX,
or whether it should remain as it is currently documented (for future
versions of NetBSD - there's no longer any thought of changing this in
any NetBSD 5 update).

Richard's primary motivation is (if I paraphrase correctly) that users
who are expecting a POSIX environment might have a ${ENV} that runs a
shell script, and if they do that they quickly get a fork bomb, as every
shell started runs ${ENV} again, ...

My argument is (basically) that it is trivial to simulate the POSIX rule
inside ${ENV} (sh(1) tells you exactly how to do it, though without
mentioning POSIX), but if the POSIX method is invoked, there's no way to
get back to the NetBSD behaviour of having ${ENV} run for every shell
(short of inventing a new environ var and making that serve this function
of course ... and even that isn't truly safe, as whatever name we invent,
someone might be using it for some other purpose, and then be surprised.)

Anyone who wants the complete rundown on the pro & con arguments should go
look in PR 42828.

What's needed now is some kind of decision, how NetBSD's shell should behave.
Before that, more (informed) opinions would be useful.

kre