Re: Patch for psql History Display on MacOSX

I wrote:
> What I'm inclined to do based on this info is to start the loop at
> history_base - 1, and ignore NULL returns until we're past history_base.

I poked at that for awhile and decided it was a bad approach. It emerges
that libedit's history_get() is just as full of version-specific
misbehaviors as the next_history() approach. While we could possibly
work around all those bugs, it's true in all libedit versions that
history_get(N) is O(N) because it iterates down the history list. So
looping over the whole history list with it is O(N^2) in the number of
history entries, which could well get painful with long history lists.

What seems better is to stick with the history_set_pos/next_history
approach, but adapt it to use previous_history when required. This is
O(N) overall in both libreadline and libedit. I therefore propose the
attached patch.

Experimenting with this, it seems to work as expected in Apple's
libedit-13 and up (corresponding to OS X Snow Leopard and newer). The
fact that it works in pre-Mavericks releases is a bit accidental, because
history_set_pos() is in fact partially broken in those releases, per
comments in the patch. And it doesn't work very well in libedit-5.1 (OS X
Tiger) because history_set_pos() is seemingly *completely* broken in that
release: it never succeeds, and we end up iterating over a subset of the
history list that does not seem to have any rhyme or reason to it.
However I don't think that this patch makes things any worse than they
were before with that release.

I only tried this directly on Tiger, Snow Leopard, and Mavericks. I
tested libedit-28 by compiling from source on a RHEL machine, so it's
possible that there's some difference between what I tested and what
Apple's really shipping. If anyone wants to try it on other platforms,
feel free.

[ wanders away wondering how it is that libedit has any following
whatsoever ... ]

regards, tom lane