[ Fixes bugs 20740, 16680, and god knows what else :) ] double the

2002-02-22  Chris Toshok  <toshok@ximian.com>

	[ Fixes bugs 20740, 16680, and god knows what else :) ]
	* gui/widgets/e-addressbook-model.c (create_card): double the
	allocated size every time we need more space instead of using a
	fixed size increment.  this helps huge queries.  Also, remove the
	gtk_object_get of "file_as", as it was dead code.
	(book_view_loaded): handle errors here (by popping up a dialog).

	* backend/pas/pas-backend-ldap.c (view_destroy): search_idle ->
	search_timeout.
	(build_card_from_entry): comment out some spew, and unref ecard
	when we're done to plug a memory leak.
	(send_pending_adds): send along to the client all the cards we've
	been saving up.
	(poll_ldap): use a timeout for ldap_result to keep the backend
	from blocking (and it turns out keep the frontend from hanging
	waiting on a ref to complete) on large db's with few matches.

	Also, add some fairly smart, self-tuning aggregating of cards.
	Keep track of the number of cards we've sent the last time through
	as well as this time, and estimate the number we want to aggregate
	the next time based on them (we average them at the moment),
	subject to maximum/minimum number of cards.  also, we have a
	maximum aggregation time, after which we force a flush if there
	are pending cards and recalculate our target pending number.
	there's a minimum wait time to possibly keep outselves from
	spamming the ui, although it's 0 at the moment.

	Lastly, make sure to only notify the GUI of status messages when
	we need to.  this results in a *huge* savings.
	(ldap_search_handler): initialize all the pending card stuff, and
	use a timeout instead of an idle function for poll_ldap.

	* backend/ebook/e-book-view-listener.c
	(e_book_view_listener_queue_response): performance optimization
	for large adds.  If we're a CardAddedEvent and there's an existing
	CardAddedEvent at the end of the queue, just concat the lists of
	cards together.  This is to keep the gui from falling further and
	further behind the ldap backend, which is merrily spewing updates
	at the gui.

svn path=/trunk/; revision=15807

1 files changed, 24 insertions, 2 deletions

diff --git a/addressbook/backend/ebook/e-book-view-listener.c b/addressbook/backend/ebook/e-book-view-listener.c
index 40e467ef93..9981375631 100644
--- a/addressbook/backend/ebook/e-book-view-listener.c
+++ b/addressbook/backend/ebook/e-book-view-listener.c
@@ -72,8 +72,30 @@ e_book_view_listener_queue_response (EBookViewListener         *listener,
 		g_free (response);
 		return;
 	}
-	
-	listener->priv->response_queue = g_list_append (listener->priv->response_queue, response);
+
+	/* a slight optimization for huge ldap queries.  if there's an
+	   existing Add response on the end of the queue, and we're an
+	   Add response, we just glom the two lists of cards
+	   together */
+	if (response->op == CardAddedEvent) {
+		GList *last = g_list_last (listener->priv->response_queue);
+		EBookViewListenerResponse *last_resp = NULL;
+
+		if (last) last_resp = last->data;
+
+		if (last_resp && last_resp->op == CardAddedEvent ) {
+			response->cards = g_list_concat (last_resp->cards, response->cards);
+			g_free (response);
+			/* there should already be a timeout since the
+			   queue isn't empty, so we'll just return
+			   here */
+			return;
+		}
+		else
+			listener->priv->response_queue = g_list_append (last, response);
+	}
+	else
+		listener->priv->response_queue = g_list_append (listener->priv->response_queue, response);
 
 	if (listener->priv->timeout_id == 0) {