GNOME Bugzilla – Bug 330406
copying entire document inside update_syntax_regions
Last modified: 2014-02-15 12:53:14 UTC
I've been doing some profiling on gedit, trying to figure out why replace-all operations are slow (bug 325853). Using the xml file from that bug, >95% of the time is spent inside the gtk_text_iter_get_slice call from update_syntax_regions, 40% of that in copy_segment. Tracing with gdb, it appears that gtk_text_iter_get_slice is duplicating the *entire* document (followed by a strlen call). It appears that for that file (with all the replace strings I tried, some simple, some complex), that the regions and table->len values were virtually always equal - which causes it to copy everything. Is line 2261 of gtksourcebuffer.c in cvs "if (region < table->len") supposed to have "<=" instead of "<" ? I don't really know anything about gtksourceview's internals, but changing that didn't seem to have any negative consequences, and made gedit's replace-all operations on the xml file take ~40 seconds instead of a few hours.
Thanks for your investigations James. However if (region < table->len) doesn't sound right to me, since we are doing g_array_index (table, SyntaxDelimiter, region) and afaics the elements in a garray go from 0 to table->len - 1.
there is something totally weird going on here: stick a printf of the GArray len in build_syntax_regions_table() /* add the delimiter to the table */ g_array_append_val (table, delim); g_print ("N: %d\n", table->len); for a normal file, for instance gtksourcebuffer.c itself, I get a len of 622, for shorter files a bit less, but for the incriminated xml file I just get 2!
(In reply to comment #2) > for a normal file, for instance gtksourcebuffer.c itself, I get a len of 622, > for shorter files a bit less, but for the incriminated xml file I just get 2! I'd been seeing 2 for the xml file too, but I hadn't looked closely enough to see if that was abnormal.
Obsolete, new engine doesn't do that.