Marc Simpson

Apologies for omitting platform information; yes, macOS (10.14.6). The failure _isn't_ the same for me under `/bin/ksh` (referenced in the original issue, perhaps it wasn't clear that this was output...

Here are two more datapoints: 1. In a fresh `ksh` (2020.0.0-beta1-166-g) session, I can establish a connection to NIST on fd 3 if it supplants an existing connection to Freenode...

Thanks for testing. Just to clarify: do you agree that this is a bug? (I see that the label was removed just before your first reply.)

Additional datapoint: crash in 93u+ with $ echo $KSH_VERSION Version AJM 93u+ 2012-08-01 $ exec 4/dev/tcp/ntp-b.nist.gov/13 Killed: 9 while this _doesn't_ crash `g1025e369`. (Note fd 4 rather than 3.)

> > Just to clarify: do you agree that this is a bug? > > Whether this issue documents a bug or long standing behavior that should be changed is...

Thanks for the explanation; this approach to prioritisation wasn't clear to me.

PS—No such crash with `set -o emacs`.

As noted on the pull request thread (#1437), while @krader1961's change seems to fix the original issue I've observed a related `ed_expand` crash in the following situation: - Launch `ksh...

@krader1961: Given that this changes (long) established behaviour and can break users' interactive sessions, can't you include this function as an _example_, and document it as a possible improvement to...

> @0branch Sure, we could do that. But that is essentially the old situation. Which meant absolutely no one knew the example functions had suffered bit rot and were not...