Sorry, but stability is more important. It was probably not a good idea
to backport all those optimization patches. From the optimization
patches merged on master, there were several other problems (not
affecting 3.18 I think, but there might be other problems not yet
discovered).

Chris Mayo authored Jun 02, 2016 and Sébastien Wilmet committed Jun 10, 2016

This caused a problem in Meld.

https://bugzilla.gnome.org/show_bug.cgi?id=767085

Move some keywords from errors to valid keywords

https://bugzilla.gnome.org/show_bug.cgi?id=765639

It is often the case that a GtkSourceMap will be packed into the widget
hierarchy, but not visible unless the user has enabled the feature.

This means that we receive a lot of signal notifications when we can't
do anything about it. So instead, toggle the signal activity based on
whether or not we are visible.

https://bugzilla.gnome.org/show_bug.cgi?id=765358

We can remove the calls to check if the cursor is visible for every visible
line by keeping track of when the property changes. This removes a
gtk_text_view_get_cursor_visible() and gtk_text_gutter_render_get_view()
for every line in the rendering process.

https://bugzilla.gnome.org/show_bug.cgi?id=765356

This makes the cost of ::query_data() for the line gutter renderer
equivalent to the cost of the g_strdup_printf() previously.

https://bugzilla.gnome.org/show_bug.cgi?id=765356

Avoid looping division to count digits. Not huge gains or anything, but
common to avoid loop+division in general. ~25% faster for numbers into
1000s.

https://bugzilla.gnome.org/show_bug.cgi?id=765356

When we simply have a subclass that overrides ::query_data vfunc and no
signals connected, we can avoid the signal emission altogether.

This is similar to what gtk_widget_draw() does to avoid signal emission.

https://bugzilla.gnome.org/show_bug.cgi?id=765300

When calculating line heights for gutter cells, we don't need precise
access to the beginning of the newline breaks. forward_to_line_end()
will place us just before the \r\n, \n, etc. Instead, we can cheat and
use the line index to jump to the next line and then backwards a char.

This is clearly a micro optimization (about 1-2% on my tests), but as
the line numbers are roughly 2x the cost of the text window, we just
need a few more of these here and there.

By connecting to the GtkWidget::draw signal, GtkSourceGutter was preventing
GtkWidget to take the fast path by avoiding signal emission. (See
gtkwidget.c for how it checks for pending signal handlers).

This instead adds a private draw function in the gutter that is called by
the view during the render case.

I've verified that this allows test-widget to hit the fast path in
gtkwidget.c.

If wrapping is disabled, then we can avoid calculating any iter locations
(which is quite expensive) since we only have a single row. The alignment
calculation will still take effect just fine.

https://bugzilla.gnome.org/show_bug.cgi?id=765104

https://bugzilla.gnome.org/show_bug.cgi?id=762592

Python 3.5 has added two new keywords to the language
defined in PEP-0492. The words are currently not
highlighted. This patch adds 'async' and 'await' to
the list of keywords.

https://bugzilla.gnome.org/show_bug.cgi?id=762026

The current lang file for vala is missing a context
for regular expressions. This patch adds the regex
definition from the javascript file which is an
existing and very mature implementation.

https://bugzilla.gnome.org/show_bug.cgi?id=697054
Fixes #697054

Template strings are part of ES6 and are currently not
highlighted as strings. This patch adds a context that
gives template strings the same color as single-quoted
and double-quoted strings. The brackets of the template
expressions are given the color of functions. The
expressions themselves are highlighted as functions
too.

https://bugzilla.gnome.org/show_bug.cgi?id=758024
Fixes #758024

Wayland requires popup windows to be attached to another toplevel
window so that it can be translated either as an xdg-popup or a
subsurface depending if there is a grab associated with the popup or
not.

GtkCompletionInfo uses gtk popup windows type but doesn't set the
transient relationship with the toplevel window, causing the
CompletionInfo window to fail to map under Wayland.

Reuse the attached_to widget to determine the toplevel window and set
the transient relationship prior to map the CompletionInfo window so
that it can work under Wayland as well.

Bugzilla: https://bugzilla.gnome.org/show_bug.cgi?id=759623

It is better, since it is the state actually used for rendering
and we want that color

https://bugzilla.gnome.org/show_bug.cgi?id=759930

gtk_render_layout() was already used. Normally it takes care of the
theme, so no need to get the foreground color with
gtk_style_context_get_color(), create manually a PangoAttribute and
apply it to the PangoLayout.

If things break, maybe calling gtk_style_context_set_state() (with
save/restore) is necessary, but it seems to work fine without it.

There is already a save/restore.

https://blogs.gnome.org/mclasen/2015/11/20/a-gtk-update/

https://blogs.gnome.org/mclasen/2015/11/20/a-gtk-update/

- Move the variable inside the function, so there is one less global
variable.
- Use G_N_ELEMENTS().
- Better function and variable names.