lean.lang: highlight attributes

See merge request GNOME/gtksourceview!281

Treat a string enclosed in slashes as a regular expression.
The name 'regexpr' is taken from the book.

Fixes #286.

 Changes to be committed:
	new file:   tests/syntax-highlighting/file.awk

awk.lang: add context for /regexpr/

Closes #286

See merge request GNOME/gtksourceview!280

Adwaita: add style for identifiers

Closes #286

See merge request GNOME/gtksourceview!282

We risk getting bounds that are out of view if we try to use various
natural words which aren't necessarily going to be the same in any syntax
language.

This tries to be simpler and just looks for non-space characters.

Be a bit more careful about clearing the hovered_at as well as ensuring
that we take the visible_rect into account when clamping the location
for the popover.

We don't use gtk_popover_popup() anywhere else, so keep symmetry by using
gtk_widget_hide() too. The only value right now that I see with the
popover variant is that it could end up cascading to the root which
isn't something we necessarily want.

If it is not active, then we want to be in something that is more akin to
a "backdrop" state and ignore the current line highlight.

One minor value of this is that we can set the name without incurring the
GMainContext/Source locks multiple times. Just set it (without those locks)
after creation so long as we have a GSource pointer.

Additionally, and more importantly, if we keep the GSource around, we can
change the ready time on the GTimeoutSource instead of remove/add flows
which are not ideal when you do as many updates as we do.

When using gtk_popover_set_autohide(..., TRUE), we incur device grabs
implicitely by GDK via GdkPopup:autohide. Those grabs can be quite
difficult to break simply using motion/crossing events.

Instead, this manages a few motion controllers (including a shared motion
controller on the GtkRoot) to track when pointer motion has left in such
a way that we want to dismiss the popover.

Fixes https://gitlab.gnome.org/GNOME/gnome-builder/-/issues/1749

Avoid a race condition if we are force disposed while the action is
processed. Doesn't happen in practice, but I'd prefer to be explicit here.

This is just a bit too obnoxious to manage alignment around, so instead
just align based on the CSS bounding box and rely on CSS for situations
where people want something different.

We need to increase the identifier by our own max id, not the max id of
the bundle to be merged.

The whole point of ensure_snippets() is that you only parse it once until
that value is cleared (which currently we do not do).

We don't need to hold onto the bundle until finalize, we already have
protections in place.

We could probably limit these if we wanted, but it's unlikely to be
worth it compared to just memcpy() as the complexity would be greater
than the memcpy costs when you have to look at every element anyway.

A better long term solution would be to have a tooltips structure that
we can reference (other than just the array) which we don't want to
possibly mutate.

We need to handle a few more cases to be certain that we've dismissed the
informative popover, particularly so that we can stop using the
GtkPopover:autohide property for it like we do for hover assistant.

With autohide turned off, we'll be able to stop forwarding keys as the
input will not be redirected to the popover.

We don't need this anymore, so just disable autohide like we do in the
GtkSourceHoverAssistant.

We still need to handle some cases better, particularly with completion
and ensuring hover assistants can break on scroll internally, but it's a
start towards that.

We had this previously, and I thought we'd be able to drop it, but it is
still needed even with autohide changes.

Rather than use bindings for this (which won't allow us to pass along the
event further) we want to both hide the assistant and pass along the event
so that systems such as vim emulation/input methods can further process it.

This follows suit of other GtkSourceAssistant by removing the grabs that
would otherwise be taken by the completion list and instead uses
GtkPopover:autohide=FAlSE.

This allows us a bit more flexibility in how events are routed but also
ensures that input doesn't continually enter/leave the source view on
every display of the list.

It also simplifies the grab management for the native surfaces since they
no longer need to track crossings and break grabs.

Otherwise we risk providers potentially not filling in the data for the
next rendering.

The baseline really seems to mess things up, so instead just try to match
the best we can so both single line and multi line look nice.

While very unlikely, it is possible that text changes programatically in
such a way that this could be possible that we need to handle the fully
different form of search text.

Instead of trying to reposition a popover synchronously from the popover's
frame-clock, try to perform that work from the beginning of the view's
next frame clock. That way, we are less likely to mess up any sort of
allocation/state used by the view.

This tries to avoid a flapping situation with completion lists where we
might get an initial notify::empty and then very shortly thereafter get
a notify::empty again with a different value.

When this would happen, we'd hide/show the completion assistant within the
same frame cycle, yet resources would have already been discarded from the
previous call to gtk_widget_hide().

If you have an active snippet (especially from a previous copmletion), then
tabbing should move you through the snippet positions, not activate the
completion you might be typing through. You can always still use enter to
activate any completion in the current position.

We are often racing to do measurements and size allocations against round
trips from the display server. Making this column able to contain an
ellipsis allows the machinery to be more forgiving in case we get too small
of an allocation inbetween roundtrips.

The hope for this originally is that we could use the result of this
function to determine what level of widget invalidation we needed on
the GtkSourceView to ensure we had an allocation for the next snapshot.

In practice, it didn't matter because either way we need to do a full
gtk_widget_queue_resize(). However, it does attempt to avoid invalidating
some state in the popover so it's useful enough to keep it around so we
know explicitely when we've changed things.

Using 0 was not really useful because it meant we could never compare the
rectangle with a previous value. gtk_popover_get_pointing_to() will always
return 1 for a width if it was set using a zero-width rect.

Using 1 means we have a stable value that can be compared across a setter
and getter for the GtkPopover.

When we update our position from the next frame cycle, we need to also
ensure that we measure/size_allocate for the updated popover. Otherwise we
risk having warnings about allocations not being up to date.

This fixes a bunch of flickering in the source view while using completion
popovers which rapidly update.

We were missing the set_property/get_property handlers for :page-size.

This optional vfunc can be used by consumers to get what the typed text
should be for a proposal. This can be useful when creating specialized
models that filter existing items such as high-priority filtering.

This adds a getter for API symmetry. It's useful to monitor the changes of
models in conjunction with an additional signals.

This signal is emitted when a model is changed for a particular provider.

This API can be used to retrieve the list of providers that are found
within a completion context. You might use this to make a filtered list
of models and prioritize some items into a high-priority model.

This helps to create some visual separation between proposals that come
from different providers. At some point, we might also want to include
the provider title (or at least provide a way to display that via a
property on GtkSourceCompletion).

Interface method which have a default implementation can be marked
virtual instead of abstract. This avoids an error for a missing
implementation of an abstract method.

vala: Mark CompletionProposal.get_typed_text as virtual

See merge request GNOME/gtksourceview!284

This is useful so that we can use a particular path when running tests out
of tree (which ensures we are testing changes) versus when running
installed tests using the installation prefix (even if that is not in the
search path).

That way we don't need to include paths for TOP_SRCDIR to get access to
the def.lang styles.

If we are running out of tree, we still want to use languages and RelaxNG
schema from the source tree. But if we do not have access to that, we can
fallback to the installation prefix (which is now handled in
GtkSourceLanguageManager).

Fixes #287

This fixes the highlighting of boolean literals.

Commands are not required to start at the beginning of a line.

lean.lang: various fixes and clean-ups

See merge request GNOME/gtksourceview!283

Leftover from conversion to final type.

This isn't used anymore, and we won't need it as we have GBindingGroup
upstream in GLib now.

We need recent GTK and GLib for this release so that we have the necessary
bug fixes we rely on.

Now that we have GSignalGroup in GObject directly, we no longer need to
carry our copy of the implementation in-tree. Just use GObject's directly.

This way things still relatively line up with the text inside the hover
assistant popover.

Using a bold red color for errors might have matched the original styles
of Solarized when it was captive to a terminal, but it does not very much
match modern IDEs and editors when it comes to what the pango renderer
can do.

Instead, use the expected error styling but color to match the Solarized
palette.

We might have to compile libsass, which will want a newer cpp_std.

Not all of these are intended to be used as installed tests, but rather
to be used by CI to ensure that merge requests don't regress features.

Fixes #288

We can compile on older releases just fine, even if it means that some
issues would be present if that were the case.

This is useful for IDEs, and needs standardization upstream at somepoint
so that we can use it across tooling.

When building on Windows with MSVC it fails to build pcre2, but the
issue was already fixed in WrapDB.

meson: Update pcre2 wrap

See merge request GNOME/gtksourceview!286

These can use error underlines instead of the blocky forms that are there
already as we've done in other schemes.

This just looks a bit better since we want to match under the whole line
not just where underbar might be.

We need to provide the length or we wont properly join the strings in all
cases here.

This requires on state within the testsuite directory, and therefore cannot
be used for system installed tests.

Fixes #291