Re: Proposed history improvements

[Zefram <zefram@xxxxxxxx>]

Wayne Davison wrote:
>                        My change makes an ignored command stick around
>just until the next command is typed.

Nice.

>                                                             Do you
>think something like this should require a separate option?

Probably, but I don't see a problem with the option being on by default.

>2. I thought it would be cool if I could hint to the history system
>about which lines I consider more important.  With HIST_DEPRECATE_SPACE
...

This bit is getting too complicated.  Instead of having multiple
competing options (HIST_DEPRECATE_SPACE and HIST_EXPIRE_DUPS_FIRST --
where's HIST_DEPRECATE_DUPS for the opposite priority ordering?), let's
have some configurable shell code to decide how important each history
entry is.  We can have a shell function that gets called whenever a
line is added to the history, which gets to reply a number saying how
important the line is.  The effect of both HIST_DEPRECATE_SPACE and
HIST_EXPIRE_DUPS_FIRST being on can then be achieved by

	function histimportance {
		if [[ -n "${(k)history[(r)$1]}" ]]; then
			REPLY=1
		elif [[ "$1" == " "* ]]; then
			REPLY=2
		else
			REPLY=3
		fi
	}

Or probably a better way would be to have a virtual associative array
giving this importance number for each history element, initialised
according to the existing options, and have a "histadded" function that
has the option to change this number for the current event or in fact
for any previous event (allowing duplicates to be expired oldest-first
instead of the newest-first that the above code does).

Other history stuff that could similarly be moved to shell code: deciding
whether a history line is to be stored at all (say, the "histadded"
function returns non-zero to say "don't add this to the history");
deciding whether a history line is to be saved in the history file
(another array, paralleling the `importance' one described above, or
an overloaded meaning for the importance array); possibly the history
file saving itself.  Actually almost the whole history mechanism *could*
be moved out of the C nd into shell code.

-zefram