This branch uses EventKit, an ultra-simple library for implementing
events. The object implementing the methods maintains its own emitter
object rather than doing a mixin like Emissary encourages. This will
make it easier for us to deprecate ::on on the object itself. Unlike
emissary, the EventKit Emitter implements a super minimalistic API that
only allows one value to be emitted and always returns a Disposable
from subscriptions.
Previously disabling a theme wouldn't wait for the full reload
to finish since and the spec would intermittently fail because
the promise was fulfilled after the spec completed and the
subscription was leaked.
The problem here is that the lines in the editor no longer reflect the
lines on screen, but we're looking for line nodes corresponding to the
editor's current contents.
This caused problems in the case of find-and-replace:select-all with
multiple editors into the same file.
* a marker is created on the TextBuffer capturing the selection
* DisplayBuffer:create-marker is fired from the first DisplayBuffer.
The marker is turned into a selection which is merged into the current
selection, deleting the marker that was created.
* DisplayBuffer::handleBufferMarkerCreated is called on the second
DisplayBuffer. The marker has been destroyed at this point, so it emits
DisplayBuffer:create-marker with undefined.
* ERROR>
Closes#3364
* Simplify scrollbar refresh and measurement by using imperative DOM
manipulation instead of React to hide/show scrollbars.
* Rename `::performInitialMeasurement` to `::becameVisible`
* Break `::checkForVisibilityChange` out of `::pollDOM` and use it in
to check for the element becoming visible in `componentWillUpdate`.
* Don't rely on stored visibility state anywhere. Always check again.
This could potentially be cached for an update cycle but being wrong
about this is disastrous so I'm being conservative.
This moves observation of the config keys to Editor, which assigns the
invisibles hash or null on the TokenizedBuffer via the DisplayBuffer to
control whether we render invisibles or not.
This commit also extends config.get to merge default values into the
returned object if both the assigned and default values are objects.
This allows 'atom.invisibles' to be treated as an object that always has
the default values filled in.
Per the discussion with @kevinsawicki in #3168, I have added a config
option `editor.confirmCheckoutHead` that defaults to `true`. When the
`editor:checkout-head-revision` command is executed, a confirmation
dialog now shows that states which file is to be reverted and asks the
user to "Revert" or "Cancel". If the config option is set to `false`,
the old behavior, simply reverting without prompting, is used.
I also added tests to ensure that the confirmation dialog is displayed
or not in the right configurations.
Check enabled names against the bundled names and ensure
one bundled UI and one bundled syntax theme are enabled,
defaulting to the dark themes when necessary.
Previously, each editor observed font-related config values on its own
and applied inline styles to honor them. This made it difficult to style
the editor like a normal element with CSS.
Moving this to a global stylesheet that targets editors via the .editor
selector means that the font size setting can be overridden in specific
contexts, such as when using mini editors.
In Chrome 31, setImmediate was yielding better behavior. But now Chrome
35 seems to behave more smoothly when we use requestAnimationFrame, and
the delay for keystrokes is non-existent.
We sample both the background color of the editor and the gutter. If the
gutter has an actual background color, we use it. Otherwise we fall back
to the same background as the editor.
This ensures lines have an opaque background that matches whatever the
editor is styled as, but avoids the need to apply the .editor-colors
class to the .lines div. That approach fell down when people were
setting the background color via means other than .editor-colors, such
as styling mini editors via the .editor.mini selector in the settings
view.
This is because of the removal of the gutter measurement. When there
was measurement, every load of the editor would measure and request at
least one render based on the reset of the gutter width. These
specs don’t need to call runSetImmediateCallbacks() as they either
don’t do anything to cause a render or they render immediately
(in the case of updated options).
In some cases, we need to make sure nothing happened, so I added a
hasSetImmediateCallbacks() function, which is used in specs where
nothing should have happened.
Previously, these measurements were always performed when the editor
component was mounted. This didn't work in situations where the
component was mounted in a non-visible state. This commit includes a
visibility check in the resize polling we were already doing, kicking
off the measurement process as soon as the editor is visible.
This doesn't seem to adversely affect the scroll experience, and it's
much simpler. I want to avoid preventing the default action of
mousewheel events if they don't actually lead to scrolling, and making
the behavior synchronous will make that a lot easier.
This makes the actions that we perform in the poll loop explicit, and
will prevent the accumulation of polling-related behavior in the
::measureScrollView method.
This allows the line height to be styled via CSS. I would actually
like to allow all these properties to be assigned via CSS rather than
explicitly via the settings view, but that can be deferred until the
old editor is removed.
