Community
Participate
Working Groups
I will provide a patch soon.
New Gerrit change created: https://git.eclipse.org/r/145490
Improvement using tests from snippet added in patch: CREATE_REVERSE_ORDER=true 1376ms -> 535ms CREATE_CTOR_2PARAM=true 4842ms -> 4108ms CREATE_REVERSE_ORDER is the most important here.
Rationale: In our software (SmartGit) we show git tags in UI. Some users have repositories with very big number of tags, one user reported 21000.
With CREATE_REVERSE_ORDER set to true, the time on my machine is ~280ms. This is down from ~4500, very nice!
Gerrit change https://git.eclipse.org/r/145490 was merged to [master]. Commit: http://git.eclipse.org/c/platform/eclipse.platform.swt.git/commit/?id=de95eb78d372a16b034c585d679aa7cfffc17c4d
Thanks for the patch Alexandr, this is a nice improvement. Can the same performance gain be made in Table? (In reply to Eclipse Genie from comment #5) > Gerrit change https://git.eclipse.org/r/145490 was merged to [master]. > Commit: > http://git.eclipse.org/c/platform/eclipse.platform.swt.git/commit/ > ?id=de95eb78d372a16b034c585d679aa7cfffc17c4d In master now.
Is there a specific problem (or even a reported bug) with Table you have in mind?
(In reply to Alexandr Miloslavskiy from comment #7) > Is there a specific problem (or even a reported bug) with Table you have in > mind? No, but Table uses the same GTK internals and in SWT the code is quite similar, so a lot of Tree fixes apply to Table as well. It's not urgent, I mention it only because if your product also uses Tables anywhere else then it's likely this optimization will apply there as well.
Verified in I20190710-0610.
What we have currently I think is no real bulk insert. It is still a one row at a time approach. According to http://scentric.net/tutorial/sec-treemodel-add-rows.html I should do 3 things on bulk insert: 3.3.3. Speed Issues when Adding a Lot of Rows A common scenario is that a model needs to be filled with a lot of rows at some point, either at start-up, or when some file is opened. An equally common scenario is that this takes an awfully long time even on powerful machines once the model contains more than a couple of thousand rows, with an exponentially decreasing rate of insertion. As already pointed out above, writing a custom model might be the best thing to do in this case. Nevertheless, there are some things you can do to work around this problem and speed things up a bit even with the stock Gtk+ models: Firstly, you should detach your list store or tree store from the tree view before doing your mass insertions, then do your insertions, and only connect your store to the tree view again when you are done with your insertions. Like this: ... model = gtk_tree_view_get_model(GTK_TREE_VIEW(view)); g_object_ref(model); /* Make sure the model stays with us after the tree view unrefs it */ gtk_tree_view_set_model(GTK_TREE_VIEW(view), NULL); /* Detach model from view */ ... insert a couple of thousand rows ... gtk_tree_view_set_model(GTK_TREE_VIEW(view), model); /* Re-attach model to view */ g_object_unref(model); ... Secondly, you should make sure that sorting is disabled while you are doing your mass insertions, otherwise your store might be resorted after each and every single row insertion, which is going to be everything but fast. Thirdly, you should not keep around a lot of tree row references if you have so many rows, because with each insertion (or removal) every single tree row reference will check whether its path needs to be updated or not. I do not see any code in swt working this way. Did anybody try this out? Is it still true or not related to latest gtk versions?
My experience with GTK is that things usually don't work the way I expect :) You're welcome to give it a try. If that gives a good performance improvement, that would definitely be nice.
