Zsh Mailing List Archive
Messages sorted by: Reverse Date, Date, Thread, Author

Re: rotating the history file



On Nov 3, 12:18am, Eric Smith wrote:
} Subject: rotating the history file
}
} however humungous history files are inelegant and I would like to
} start rotating the file every few thousand entries or so.  Like log
} files with log.1 .. log.2 .. log.3 .. log.n

What would you eventually want to do with the old files?

There are a whole lot of complications I can think of with attempting to
rotate "every few thousand entries," involving interactions of multiple
shells, the way history events expire when the runtime history exceeds
$HISTSIZE and when the file size exceeds $SAVEHIST, and the unpredictable
number of events that may occur in any given interactive session.  The
conclusion I've come to after several failed attempts to explain all of
it is that the only way to handle this request is to try to prevent the
history from filling up, so that events normally leave the history only
when you rotate the files.

That means setting SAVEHIST much bigger than HISTSIZE, but then never
letting the HISTFILE reach its maximum size.  How much bigger SAVEHIST
needs to be depends on how many shells you run at once, how different the
command histories are, and how many new commands you tend to run in any
session.

As I wrote back in August:

>When zsh starts up, reading the history file is the very last thing that it
>does before entering the main interaction loop.
>
>When zsh exits normally, it first writes the history, then reads .zlogout,
>and finally runs the traps.

So you probably want to rotate the files during shell startup, so that only
one shell performs the rotation, and then if you exit from multiple shells
their history is merged and used as the basis for the next rotation when
another shell starts up.  (I hope that made sense.)  You'll need to have
either append_history or inc_append_history set; things get more complex
if you also want share_history, so I'm going to ignore that for now.

This will all work best in 3.1.9 and later, where hist_save_no_dups can
be set to avoid saving multiple copies of the same command in HISTFILE.
Otherwise, if you do run multiple shells, you may find the file to be
rotating more often than you like.

Begin by defining a couple of variables:

    ROTATEHIST=2000	# How many lines of growth before rotating
    MAXHISTFILES=20	# However many rotations you want to keep

Then rotating the file is pretty easy with `zmv':

    zmv -Qf "($HISTFILE).(<->)(On)" '$1.$[$2+1]'
    mv -f $HISTFILE $HISTFILE.0
    rm -f $HISTFILE.$MAXHISTFILES

The `(On)' qualifier (and hence -Q option) in the zmv command causes the
file names to be ordered such that each will move out of the way before
the next is renamed to replace it.  The -f is necessary because zmv
checks the existence of all destination files before it actually moves
any of them, so it can't tell that this ordering is safe.

The tricky part is deciding when to rotate the files.  The most accurate
way is simply to read in the history and see how many entries you end up
with.  So:

    rotate_history() {
	autoload -U zmv
	zmodload -i zsh/parameter || return 1

	# Reset HISTSIZE to SAVEHIST so we can read the entire file
	local histsize=$HISTSIZE HISTSIZE=$SAVEHIST

	# Read the file and see how many entries we ended up with
	fc -R
	if (( ${#${(k)history}} >= histsize + ROTATEHIST ))
	then
	    # Rotate the files if we got this far
	    zmv -Qf "($HISTFILE).(<->)(On)" '$1.$[$2+1]' || return 1
	    mv -f $HISTFILE $HISTFILE.0 || return 1
	    rm -f $HISTFILE.$MAXHISTFILES

	    # Restore HISTSIZE to truncate the history, and write it out
	    HISTSIZE=$histsize
	    fc -W
	fi

	# Empty the history completely so it can be re-read later
	HISTSIZE=0	# This is only safe from .zshrc or .zlogin!
    }

This assures that any time the HISTFILE is rotated, it's left with no
more than HISTSIZE entries.  Using SAVEHIST larger than HISTSIZE in
combination with append_history will assure that multiple shells can
still merge their histories into the file without losing entries; but
as no shell will retain more than HISTSIZE entries in any case, that's
a safe size to leave HISTFILE after rotating.

The final HISTSIZE=0 is to prevent the newly-rewritten history from
being merged with the copy that was read by `fc -R'.  I.e., you can't
prevent zsh from reading $HISTFILE just before it starts the main loop,
so to avoid duplication the history has to be empty at that point.
You could instead use `setopt hist_ignore_all_dups' but that's a bit
less efficient.  Either way, a side-effect of this scheme is that your
history event numbers (e.g. in your prompt) may be much larger than the
actual number of history entries loaded.

-- 
Bart Schaefer                                 Brass Lantern Enterprises
http://www.well.com/user/barts              http://www.brasslantern.com

Zsh: http://www.zsh.org | PHPerl Project: http://phperl.sourceforge.net   



Messages sorted by: Reverse Date, Date, Thread, Author