* Improving resizehistents()
@ 2002-02-13 2:00 Wayne Davison
2002-02-13 17:14 ` Bart Schaefer
0 siblings, 1 reply; 4+ messages in thread
From: Wayne Davison @ 2002-02-13 2:00 UTC (permalink / raw)
To: Zsh Workers
While I was looking at the HISTSIZE stuff, I noticed that the
resizehistents() function does not respect the HIST_EXPIRE_DUPS_FIRST
option. I decided to break out the code that deals with this into a
separate function that could be used by both prepnexthistent() and
resizehistents(). Here's the result.
Optimization note: when the new putoldhistentryontop() function is
called in a loop with the HIST_EXPIRE_DUPS_FIRST option set, it is
possible that we could scan the start of the history list once for
every line we remove. While I could complicate the function and
make it possible for the scan to continue where it left off, I
decided to leave it simple (and inefficient). I think this is the
right choice since resizehistents() doesn't get called a lot. If
anyone disagrees, let me know.
..wayne..
---8<------8<------8<------8<---cut here--->8------>8------>8------>8---
Index: Src/hist.c
--- Src/hist.c 6 Feb 2002 16:48:28 -0000 1.37
+++ Src/hist.c 12 Feb 2002 22:55:46 -0000
@@ -913,11 +913,34 @@
return he;
}
+static void
+putoldhistentryontop(void)
+{
+ Histent he = hist_ring->down;
+ if (isset(HISTEXPIREDUPSFIRST) && !(he->flags & HIST_DUP)) {
+ int max_unique_ct = getiparam("SAVEHIST");
+ do {
+ if (max_unique_ct-- <= 0) {
+ he = hist_ring->down;
+ break;
+ }
+ he = he->down;
+ } while (he != hist_ring->down && !(he->flags & HIST_DUP));
+ if (he != hist_ring->down) {
+ he->up->down = he->down;
+ he->down->up = he->up;
+ he->up = hist_ring;
+ he->down = hist_ring->down;
+ hist_ring->down = he->down->up = he;
+ }
+ }
+ hist_ring = he;
+}
+
/**/
Histent
prepnexthistent(void)
{
- Histent he;
int curline_in_ring = hist_ring == &curline;
if (curline_in_ring)
@@ -928,7 +951,7 @@
}
if (histlinect < histsiz) {
- he = (Histent)zcalloc(sizeof *he);
+ Histent he = (Histent)zcalloc(sizeof *he);
if (!hist_ring)
hist_ring = he->up = he->down = he;
else {
@@ -940,30 +963,13 @@
histlinect++;
}
else {
- he = hist_ring->down;
- if (isset(HISTEXPIREDUPSFIRST) && !(he->flags & HIST_DUP)) {
- int max_unique_ct = getiparam("SAVEHIST");
- do {
- if (max_unique_ct-- <= 0) {
- he = hist_ring->down;
- break;
- }
- he = he->down;
- } while (he != hist_ring->down && !(he->flags & HIST_DUP));
- if (he != hist_ring->down) {
- he->up->down = he->down;
- he->down->up = he->up;
- he->up = hist_ring;
- he->down = hist_ring->down;
- hist_ring->down = he->down->up = he;
- }
- }
- freehistdata(hist_ring = he, 0);
+ putoldhistentryontop();
+ freehistdata(hist_ring, 0);
}
- he->histnum = ++curhist;
+ hist_ring->histnum = ++curhist;
if (curline_in_ring)
linkcurline();
- return he;
+ return hist_ring;
}
/* A helper function for hend() */
@@ -1756,8 +1762,10 @@
void
resizehistents(void)
{
- while (histlinect > histsiz)
- freehistnode((HashNode)hist_ring->down);
+ while (histlinect > histsiz) {
+ putoldhistentryontop();
+ freehistnode((HashNode)hist_ring);
+ }
}
/* Remember the last line in the history file so we can find it again. */
---8<------8<------8<------8<---cut here--->8------>8------>8------>8---
^ permalink raw reply [flat|nested] 4+ messages in thread
* Re: Improving resizehistents()
2002-02-13 2:00 Improving resizehistents() Wayne Davison
@ 2002-02-13 17:14 ` Bart Schaefer
2002-02-13 18:36 ` Wayne Davison
0 siblings, 1 reply; 4+ messages in thread
From: Bart Schaefer @ 2002-02-13 17:14 UTC (permalink / raw)
To: Zsh Workers
On Feb 12, 6:00pm, Wayne Davison wrote:
}
[...]
} possible that we could scan the start of the history list once for
} every line we remove. While I could complicate the function and
} make it possible for the scan to continue where it left off, I
} decided to leave it simple (and inefficient).
My intuition is that you've made the right choice, but I know there are
some people out there using HISTSIZE into the many thousands. How big
can the history get before this creates a noticeable delay?
--
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
^ permalink raw reply [flat|nested] 4+ messages in thread
* Re: Improving resizehistents()
2002-02-13 17:14 ` Bart Schaefer
@ 2002-02-13 18:36 ` Wayne Davison
2002-02-13 19:32 ` Wayne Davison
0 siblings, 1 reply; 4+ messages in thread
From: Wayne Davison @ 2002-02-13 18:36 UTC (permalink / raw)
To: Bart Schaefer; +Cc: Zsh Workers
On Wed, 13 Feb 2002, Bart Schaefer wrote:
> My intuition is that you've made the right choice, but I know there are
> some people out there using HISTSIZE into the many thousands. How big
> can the history get before this creates a noticeable delay?
I tried the following test with $n set to 10000, 20000, and 100000:
setopt histexpiredupsfirst
HISTSIZE=$n
SAVEHIST=$n
fc -R unique$n
date ; HISTSIZE=1 ; date
(Of course, the file unique10000 had 10000 unique commands in it, etc.)
The results for the current code on my PII 433 are:
10000: 4 secs
20000: 31 secs
100000: [killed zsh after over 7 minutes of 98% cpu usage]
I then tested a function that I wrote that allows restarting the scan:
10000: 0 secs
20000: 0 secs
100000: 0 secs
I've appended a diff between my first patch and the new code. It's a
bit more complex, but not too bad. I think I'll commit it after a
little more testing. Feel free to let me know what you think.
..wayne..
---8<------8<------8<------8<---cut here--->8------>8------>8------>8---
Index: Src/hist.c
--- Src/hist.c 13 Feb 2002 18:13:14 -0000 1.38
+++ Src/hist.c 13 Feb 2002 18:27:48 -0000
@@ -914,17 +914,23 @@
}
static void
-putoldhistentryontop(void)
+putoldhistentryontop(short keep_going)
{
- Histent he = hist_ring->down;
+ static Histent next = NULL;
+ Histent he = keep_going? next : hist_ring->down;
if (isset(HISTEXPIREDUPSFIRST) && !(he->flags & HIST_DUP)) {
- int max_unique_ct = getiparam("SAVEHIST");
+ static int max_unique_ct = 0;
+ if (!keep_going)
+ max_unique_ct = getiparam("SAVEHIST");
+ next = he->down;
do {
if (max_unique_ct-- <= 0) {
+ max_unique_ct = 1; /* On next keep_going, scan 1 more entry */
he = hist_ring->down;
break;
}
- he = he->down;
+ he = next;
+ next = he->down;
} while (he != hist_ring->down && !(he->flags & HIST_DUP));
if (he != hist_ring->down) {
he->up->down = he->down;
@@ -963,7 +969,7 @@
histlinect++;
}
else {
- putoldhistentryontop();
+ putoldhistentryontop(0);
freehistdata(hist_ring, 0);
}
hist_ring->histnum = ++curhist;
@@ -1762,9 +1768,13 @@
void
resizehistents(void)
{
- while (histlinect > histsiz) {
- putoldhistentryontop();
+ if (histlinect > histsiz) {
+ putoldhistentryontop(0);
freehistnode((HashNode)hist_ring);
+ while (histlinect > histsiz) {
+ putoldhistentryontop(1);
+ freehistnode((HashNode)hist_ring);
+ }
}
}
---8<------8<------8<------8<---cut here--->8------>8------>8------>8---
^ permalink raw reply [flat|nested] 4+ messages in thread
* Re: Improving resizehistents()
2002-02-13 18:36 ` Wayne Davison
@ 2002-02-13 19:32 ` Wayne Davison
0 siblings, 0 replies; 4+ messages in thread
From: Wayne Davison @ 2002-02-13 19:32 UTC (permalink / raw)
To: Zsh Workers
On Wed, 13 Feb 2002, Wayne Davison wrote:
> I've appended a diff between my first patch and the new code.
Yeesh. That function was totally wacked except when there were no dups.
This version appears to work both correctly and quickly.
..wayne..
---8<------8<------8<------8<---cut here--->8------>8------>8------>8---
Index: Src/hist.c
--- Src/hist.c 13 Feb 2002 18:13:14 -0000 1.38
+++ Src/hist.c 13 Feb 2002 19:21:26 -0000
@@ -914,25 +914,31 @@
}
static void
-putoldhistentryontop(void)
+putoldhistentryontop(short keep_going)
{
- Histent he = hist_ring->down;
+ static Histent next = NULL;
+ Histent he = keep_going? next : hist_ring->down;
+ next = he->down;
if (isset(HISTEXPIREDUPSFIRST) && !(he->flags & HIST_DUP)) {
- int max_unique_ct = getiparam("SAVEHIST");
+ static int max_unique_ct = 0;
+ if (!keep_going)
+ max_unique_ct = getiparam("SAVEHIST");
do {
if (max_unique_ct-- <= 0) {
+ max_unique_ct = 0;
he = hist_ring->down;
break;
}
- he = he->down;
+ he = next;
+ next = he->down;
} while (he != hist_ring->down && !(he->flags & HIST_DUP));
- if (he != hist_ring->down) {
- he->up->down = he->down;
- he->down->up = he->up;
- he->up = hist_ring;
- he->down = hist_ring->down;
- hist_ring->down = he->down->up = he;
- }
+ }
+ if (he != hist_ring->down) {
+ he->up->down = he->down;
+ he->down->up = he->up;
+ he->up = hist_ring;
+ he->down = hist_ring->down;
+ hist_ring->down = he->down->up = he;
}
hist_ring = he;
}
@@ -963,7 +969,7 @@
histlinect++;
}
else {
- putoldhistentryontop();
+ putoldhistentryontop(0);
freehistdata(hist_ring, 0);
}
hist_ring->histnum = ++curhist;
@@ -1762,9 +1768,13 @@
void
resizehistents(void)
{
- while (histlinect > histsiz) {
- putoldhistentryontop();
+ if (histlinect > histsiz) {
+ putoldhistentryontop(0);
freehistnode((HashNode)hist_ring);
+ while (histlinect > histsiz) {
+ putoldhistentryontop(1);
+ freehistnode((HashNode)hist_ring);
+ }
}
}
---8<------8<------8<------8<---cut here--->8------>8------>8------>8---
^ permalink raw reply [flat|nested] 4+ messages in thread
end of thread, other threads:[~2002-02-13 19:33 UTC | newest]
Thread overview: 4+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2002-02-13 2:00 Improving resizehistents() Wayne Davison
2002-02-13 17:14 ` Bart Schaefer
2002-02-13 18:36 ` Wayne Davison
2002-02-13 19:32 ` Wayne Davison
Code repositories for project(s) associated with this public inbox
https://git.vuxu.org/mirror/zsh/
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).