The requiring of a package's main module is now decoupled from package
activation. Non-deferred packages will always be required before the
panes are deserialized. This allows the package to register any
deserializers for objects displayed in the panes.
Deferred packages can contain a 'deferredDeserializers' array in their
package.cson. If we attempt to deserialize an object with a deserializer
in the list, the package's main module will be required first so it has
a chance to register the deserializer. But the package still won't be
activated until an activation event occurs.
We may want to add an additional optional hook called 'load' which is
called at require time. We would not guarantee that the rootView
global would exist, but we could give the package a chance to register
deserializers etc. For now, registering deserializers is a side-effect
of requiring the package.
Modal dialogs can be presented while other modal dialogs are already
being displayed. Previously, dialogs were always displayed in the order
they were requested. But say you have two untitled buffers in a
pane and you close all items… You'll display prompt dialogs for both
buffers asking the user if they want to save. If the user answers yes
to the first dialog, they should see the path selection dialog before
they see the save prompt for the second buffer.
This commit uses a stack of queues to store deferred dialogs and allow
dialogs presented by the dismissal of another dialog to take precedence
over other pending dialogs.
Keeping the shutdown state as a local var in window.coffee causes spec failures because window.shutdown can only be called once in the entire spec suite
In additional, rename `registerViewClass(es)` to `registerDeserializer(s)`.
This moves us to a situation where any kind of object may want to be
deserialized, not just views.
Allowing root view to be focused was stealing focus away from the
editor whenever a click event made it to the root view. This unnecessary
switching of focus was interfering with the ability to drag tabs.
But if RootView can't be focused, focus ends up being returned to the
document body when there are no focusable elements. This would be fine,
except for the fact that we frequently bind global events on root view,
and so they aren't triggered when events are triggered on the body. We
could just bind all global events on the body, but this would require
us to always attach elements to the DOM during specs, which is a serious
performance killer in specs.
The workaround is in the keymap. When the keymap handles a key event
that was triggered on the body, it triggers the corresponding semantic
event on the root view anyway, so from the event perspective, it's as
if the root view actually had focus. The only place this might fall
down is if someone wants to capture raw key events. But that's the
keymap's job anyway, and we maybe add a hook on the keymap if such a
need ever arises.
Also, add a spec to cover the loading of keymaps in `atom-spec` and
reset the `keymap`'s internal data after each spec gets run to prevent
test pollution with keymaps.
TextMatePackage is only designed to load resources out of a TextMate
bundle. It's used only at load time, and from that point out we only
refer to our own global `syntax` data structure to access the data that
it loads.
This simplifies the loading of TextMate bundles in the spec and benchmark helpers. Since `loadBundle` was already implemented on `atom`, it made sense to move this logic here. Config is now more focused on its core job of handling configuration, not loading bundles.
The goal is that `loadPackage` will be the go-to place for loading all kinds of resources out of directories. `requireExtension` was only designed to load and activate extension modules.
The `config` object no longer stores config properties directly. Instead it stores them on an internal `settings` object, which makes it easier to serialize settings without getting them mixed up with non-setting state on the `config` object.
This replaces the old functionality of ensuring no files or directories have subscriptions in javascript. We allow this now, but we just don't allow leaked watches at the native layer.
Path watching resumes once the file is saved again. This commit allows files to be created for as-yet nonexistent paths. We won't call `$native.watchPath` until we have at least 1 subscription to the file in JS and the file exists on disk.
Also, we moved execution of the path watcher callbacks until after the callbacks data structure is updated in order to avoid confusing behavior in specs.
I added overflow hidden on the html and body tags in a previous commit to prevent rubber banding when scrolling the editor. But it broke the ability to scroll specs. This fixes that.
When we actually want to attach the root view in window-bootstrap.coffee, we call `window.attachRootView(path)` instead of calling `window.startup(path)`. Having `startup` called automatically means we can be sure any code we add there runs in every environment (including benchmark and specs). This is where we do things like setup the global keymap, parse text mate bundles and themes, and establish the window close handler. Any globals other than the root view that we want to be available in all environments should be established here. Right now that's just the keymap, but soon I want to add a global pasteboard.
Flexbox was causing layouts and repaints to cover the entire scroll view instead of just the edited line. This cuts down on DOM manipulation cost significantly.
We do this to ensure that the lines aren't longer than the scroll view if they don't have to be. We really should use min-width instead because it's automatic. Also, trigger window resize when we make the editor narrower.
Move the saving of serialized root view data to window.coffee. The window.startup method looks for window state on the atom object and instantiates the root view with that if it is present.
Introduce Point and Range objects. Selection.selectRight places an
anchor object before moving right if no anchor yet exists. Still no
visual treatment.