Angband spends too much time updating the item list.

[morth]

Well, I accidently overwrote my savefile with my testing svn version, so playing r1444 now... I noticed running is _much_ slower than it was in 3.1.0 beta, even after I disable the pricing log. Almost not playable on my 2GHz macbook pro. Profiling gives the following traceback:

Depth Image Samples Function

24 angband 1034 slay_power 23 angband 1034 object_power 22 angband 1034 object_value_real 21 angband 1034 object_value 20 angband 1034 display_itemlist 19 angband 1034 update_itemlist_subwindow 18 angband 1034 game_event_dispatch 17 angband 1034 redraw_stuff 16 angband 644 process_player 15 angband 644 dungeon 14 angband 644 play_game 13 angband 644 AngbandGame?

It seems a bit obsessive, the item list doesn't need updating at all while I'm not using items from the floor or a monster picks them up/destroys them (this was from running/resting only)

[d_m]

This is strange, because AFAIK the itemlist code hasn't been updated between 3.1.0 and HEAD. I know that as of its inclusion the itemlist code was redrawing too frequently, but that's always been the case. Thus, I am pretty sure it's not directly responsible for the slowdown you saw.

That said, it is worth cleaning up.

Anyway, I will try to see what (if anything) has changed in the itemlist drawing code and report back.

[anonymous]

If you wouldn't mind backing up your savefile, then running 3.1.0-beta again and returning some profiling data from that, it would be helpful.

I compared SVN-1098 (the point where 3.1.0 was made) against HEAD. I didn't notice any changes to object/obj-util.c that affect the itemlist. Of course, it's possible some other code did increase the redrawing frequency. Anyway, it would help to be sure that it is happening more under HEAD than 1098.

[d_m]

That last comment was by me but I forgot to sign it. Sorry.

[morth]

Maybe it's drawn just as often, but takes much longer time to do it now. Isn't the call to slay_power new? Since every sample is in that function (sorry for the messed up display), it should be that function that is slow... That said, it could probably be fixed either by optimizing the function or by calling it less.

Looking a bit at the code, the cache used in slay_power isn't very effective since it's a linear search. It should be sorted or a hash. I haven't verified that that is the actual problem though.

[morth]

I checked with 3.1.0-beta and display_itemlist is still at the top of the list, but it's way faster.

[Taha]

This sounds like the same issue documented here: http://angband.oook.cz/forum/showthread.php?t=1951

It definitely was not a problem in 3.10, and has been in all the nightlies I have tried since. I haven't done any detailed followup, except to confirm that the same savefile has the issue in the nightlies but not in 3.10. I have just had to keep the item list turned off since then.

[sorear]

There are several aspects to this.

object_value does too much; most of the complicated calculation is not needed if the result is just being compared to 0. It should be split out into object_is_worthless and object_value.

The item list is redisplayed every turn.

The item list is sorted using bubble sort, with several calls to object_value in the comparison.

Fixing any one of these would very likely make the game playable again. All three would probably be best (come on, bubble sort??).

[magnate]

r1477 removed the logging of power calculations when object_value is called from sorting the item list (power is now only logged whilst in stores, and during randart generation). Is someone who uses the item list window deems that the game is now playable again, please close this ticket.

[magnate]

http://angband.oook.cz/forum/showthread.php?t=2088

The slowdown is definitely due to the additional calls to the object power function via object_value. My plan is to avoid calling object_value when ordering lists of items, and using k_ptr->cost or e_ptr->cost instead. This will leave the ordering essentially unchanged, and has the added bonus of preventing wands and staves swapping places as you use charges.

[d_m]

Patch which uses k_ptr->cost instead of object_value()

[d_m]

I'm going to go ahead and email the Angband list to try to get this patch in sooner rather than later.

[magnate]

r1500