Changeset 5c90890 for irc_user.c

Timestamp:

2018-07-12T08:54:12Z (6 years ago)

Author:

dequis <dx@…>

Branches:

master

Children:

c82e4ca

Parents:

b9d1fdc

git-author:

dequis <dx@…> (03-07-18 06:17:17)

git-committer:

dequis <dx@…> (12-07-18 08:54:12)

Message:

Stop using the irc->users linked list, use the hash table instead

irc_user_new() mentions that the reason this list is kept is for easy
iteration. Luckily, this is the future, and GHashTableIter exists now.

The main point of this is to get rid of the g_slist_insert_sorted() in
irc_user_set_nick() which is a particularly slow part of loading large
user lists, and scales poorly

In a test with discord, the GUILD_SYNC event is now 4 times faster, on
top of the improvements of the other bee_user hash tables patch.
Combining both patches it went from 136 to 6 seconds for 50k members.

File:

• irc_user.c (modified) (4 diffs)

irc_user.c

@@ -37 +37 @@
         iu->key = g_strdup(nick);
         nick_lc(irc, iu->key);
-        /* Using the hash table for speed and irc->users for easy iteration
-           through the list (since the GLib API doesn't have anything sane
-           for that.) */
         g_hash_table_insert(irc->nick_user_hash, iu->key, iu);
-        irc->users = g_slist_insert_sorted(irc->users, iu, irc_user_cmp);
 
         return iu;
@@ -87 +83 @@
         irc_user_quit(iu, msg);
 
-        irc->users = g_slist_remove(irc->users, iu);
         g_hash_table_remove(irc->nick_user_hash, iu->key);
 
@@ -148 +143 @@
         }
 
-        irc->users = g_slist_remove(irc->users, iu);
         g_hash_table_remove(irc->nick_user_hash, iu->key);
 
@@ -175 +169 @@
         iu->key = g_strdup(key);
         g_hash_table_insert(irc->nick_user_hash, iu->key, iu);
-        irc->users = g_slist_insert_sorted(irc->users, iu, irc_user_cmp);
 
         if (iu == irc->user) {