Bold color, if unset, now reports the foreground color. Unsupported
underline/blink/reverse/italic colors also report the foreground color.

Fixes #2768

[u]intmax_t is a bit useless anyway, and not supported by
introspection.

... to allow querying the name of an alias termprop's target.

emplace_back() makes clang++ 15.0.7 error out here, for some reason.
Changing to use push_back() works, so go with that.

Fixes: #2770

We do not want to dereference if we've reached the end of the chunk.
This matches the bounds check above.

Fixes: https://gitlab.gnome.org/GNOME/vte/-/issues/2769

... and support up to 15 mouse buttons.

Fixes #2765

Zander Brown authored Mar 19, 2024 and Christian Persch committed Mar 19, 2024

Fixes: #2763

On gtk3, painting from the minifont cache seems to be slower than
painting the glyph in the first place, so don't use a cache.

Fixes #2757

Carlos Garnacho authored Mar 13, 2024 and Christian Persch committed Mar 13, 2024

Currently, make it mutually exclusive with having started a selection,
other heuristics might be added here (e.g. making it also mutually exclusive
with unselecting text, or not issuing the call for cases where text input
is known to be ignored.

#2630

Carlos Garnacho authored Mar 13, 2024 and Christian Persch committed Mar 13, 2024

This method makes the GtkIMContext request an OSK to be activated.
Currently supported on the Wayland platform.

#2630

To simplify the code, remove draw_text, move its debugging
into the draw_text_internal function, and rename draw_text_internal
back to draw_text.

Since `WITH_A11Y` is always defined, a `defined()` check leads to link
errors when `WITH_A11Y == 0`.

Likewise `GTK_ACCESSIBLE_ROLE_TERMINAL` was only introduced in 4.13.9,
thus `WITH_A11Y == 1` can fail despite passing the configure check for
4.13.

See: gtk@64316ab1
Fix: c8d7967f

Christian Hergert authored Mar 02, 2024 and Christian Persch committed Mar 02, 2024

#2747

This fixes a regression from commit
5a242c23.

Fixes: #2752

... and add loading/saving options from/to a key file.

This is an initial implementaiton of GtkAccessibleText which was added
to GTK for 4.14. It attempts to implement things in a very similar
fashion to the previous code for GTK 3 although considerable effort was
made to simplify and improve readability as to how it works.

Currently, this supports reading back what you type and what has changed
on screen. It is not yet 1:1 what the GTK 3 a11y implementation did
because ATK was doing many other things (including proxying keyboard
keys) to the other side of the a11y bus. That appears to improve
readback by screen readers in the form of "backspace" and what character
was deleted.

I expect things to get closer to 1:1 but that work is going to have to
be done inside of GTK itself first and should not require much if
anything here.

A new VteTerminal:enable-a11y feature flag property has been added
because I'm concerned about enabling this by default until the a11y bus
learns to be more lazy. Currently there is no way to "do nothing" until
a peer (e.g. screenreader) is interested in the contents.

Ideally, we would have a short-circuit like is currently implemented
by checking vte_terminal_get_enable_a11y() to avoid any sort of
contents calculation when there are no a11y observers.

It also allows disabling the GTK 3 a11y implementation just to keep
some symmetry between the APIs.

Currently, this does not implement "text-scrolled" like the GTK 3
implementation does as I'm not sure yet if there is a benefit.

Fixes: #2753

Fixes #2749

This is called frequently during Terminal::draw() and reduces overhead
there on release builds a decent amount. I didn't quanitfy exactly without
bidi patches applied, but it approaches double digits.

These are called frequently during drawing. Inlining the fast path is
extremely useful here to keep that overhead low.

Hoist it into the caller for roughly a 4.5% reduction in drawing time
using scrolling a large file in vim as the test case.

This function appears to be called a lot, especially during high scrolling
scenarios (such as ctrl+f/ctrl+b in vim). There is a typical fast path
which is that ring is updated and there is no selection. Hoist that into
the callers and leave the selection case behind a function call.

This shaves about 1.5% of application samples off time spent in
Terminal::draw().

This gets called extremly often so ensuring that we keep the common path
tight is important. This is good for about 2% of application samples
in the case of scrolling with ctrl+f/ctrl+b in vim.

This is called very frequently.

The invisible check, which should be unnecessary, is about 2.8% of
application samples in a tight scrolling loop with vim doing
ctrl+f/ctrl+b.

This reverts commit caf8a6a7.

Also bump GLib requirement to include their fix, and add an API comment
discouraging the use of large scrollback sizes.

Fixes #2504

Marko Mäkelä authored Feb 24, 2024 and Egmont Koblinger committed Feb 24, 2024

Followup of commit e1481056.

Bug-Debian: https://bugs.debian.org/1063941

#2743

If we fail to get updates from the window manager and/or compositor, then
fallback to a 10hz scheduler that will ensure we still make progress on
incoming PTY data.

This fixes a situation where the GdkFrameClock will not advance while the
window is hidden in GNOME Shell using Super+H.

#2743

#2665 (comment 2022380)