GitLab

When getting the displayed text of something which claims to implement
the accessible text interface but then doesn't, we fall back on the
name. That doesn't make sense for a spin button in general. In addition,
there seems to be a bug in Chrome/Chromium in which spin buttons which
display text fail to expose that text via the text interface. Falling
back on the name prevents other checks (e.g. valuetext property) from
happening. So this change in Orca will work around that issue in
Chrome/Chromium.

Chrome (and possibly other browsers) emits caret-moved events prior to
focus events with inputs. This can cause us to quietly set the focus
and subsequently ignore the focus event. We were already handling that
ordering for inputs exposed as entries. But inputs with a type of
number get exposed as spin buttons. This commit fixes the issue for
numeric inputs.

Python 3.10 removed aliases to Collections Abstract Base Classes that
were deprecated in Python 3.3.

Fix this by just using collections.abc directly without using the alias.

https://docs.python.org/3.10/whatsnew/changelog.html#python-3-10-0-alpha-5

The value of hyperlink.startIndex should match the offset in the text
implementation's character offset for the associated child object.
When this fails, due to a bug in the browser implementation, we can
get stuck/loop. Therefore, check for this mismatch, spew out plenty
of debugging details, and try to recover by using the hyperlink's
reported range instead of a character offset when we've gotten stuck
on a line.

If an ARIA tooltip claims to be focusable, we treat it as something
presentable even if it has no size and no content. This leads to
chattiness, and occasionally even setting the caret in, a useless
object. The latter can trigger focus events, along with object
creation and destruction that leads to looping within the content.
Solve both problems by not allowing an empty tooltip to have the
caret context.

Also fix a traceback checking for the name of a recently-destroyed
ARIA tooltip.

In LibreOffice, when we get a text-selection-changed event from an
object which claims to be focused but is not the locusOfFocus, we
update the locusOfFocus before passing the event onto the default
script. We do this because LibreOffice isn't super reliable about
emitting focus events, and the default script will ignore the event
otherwise.

Unfortunately, Calc is emitting text-selection-changed on a destroyed
object which claims to be focused. Therefore, check if this object is
dead and return early if it is.

After saving a document and returning to the Calc spreadsheet, there
is no object:state-changed:focused event for the table. As a result,
AT-SPI2 does not update its cache and we're given active-descendant-
changed events on an object which claims to be unfocused. Thus we
ignore them.

Therefore, check in the app script for this condition and clear the
cache on the table before handing the event off to the default script.

self._default_voice_name (i.e. "<synth_name> default voice") is just
the end-user (translated, even!) label for asking for the default
speech-dispatcher voice. We should not be asking speech-dispatcher a voice
with that name, and just rely on the selection from the locale.

They are like text-changed events, and (at least) Firefox can be super
spammy here.

We sometimes get event floods from Firefox during a page load. Therefore,
add document:load-complete and object:state-changed:busy false as an
indicator that it's a good time to prune events during a flood.

We ignore text-changed events during event floods, but were not checking
for the "system" suffix that Firefox uses depending on the change type.
This commit causes those events to be ignored as well.

Toolbars were being treated as 'stopAtRoles' when generating new
ancestors. We want to present that a toolbar has been entered, so
move it to the list of 'stopAfterRoles'.

If we're in a toolbar which is descendable (e.g. because we're in browse
mode), the line contents should be for the descendants and not include
the toolbar itself.

Including it as part of the contents can lead to our getting stuck when
we try to navigate past it due to a to-be-debugged issue. However, since
the toolbar itself shouldn't be on the line anyway, remove it.

Orca's loadUserSettings calls the Mouse Reviewer's activation method.
That method should check if we have Wnck and return early if we don't.

Apparently in some environments getting the pointer from the default seat
of the GDK display returns null rather than errors out. Handle this error.

In order to be performant, onChildrenRemoved returns early when the event
comes from a live region. However, if a user is navigating in live region
content and lands on something which disappears, we still want to try and
recovery rather than risk looping to the top of the page.

Cached information is associated with the top-level document. The
document associated with the event.source triggering a need to dump
cache might be nested.