The Replace All button’s enabled state is based on count of matches (via bindings), so unbinding would leave it possibly enabled when it should have been disabled.
This is to avoid an exception when the code is used with menu items that have no represented object (which it normally wouldn’t in shipping code, but it seems it should be supported nonetheless).
The system can be updating the menu only to resolve a key equivalent, hence the image is not required. The document icon can be a little expensive to create because of the SCM badge. This would be noticeable when doing a folder search with hundres or thousands of results and then pressing ⌘1, here it would previously need to create all the document icons before responding to the key.
Previously searching for something with a trailing newline would have the pop-out animation (image) include an extra empty line.
An issue is when only searching for a newline, currently this results in an empty image, so we need to special-case that.
When a menu item has a custom app shortcut it is necessary to keep the original title as we would otherwise lose the custom app shortcut defined by the user (using the original title).
We should probably update the directory if the project’s project folder is changed, and we may also consider storing this in the session, at least if we set the directory during “New Tab” (based on current file browser selection).
When batch-opening documents we only load the selected one. Previously this meant that e.g. after session restore, using Find in Open Files would not search documents in background tabs (that had not yet been clicked).
Since we do not load documents before the user selects them, e.g. after session restore or when opening multiple documents, it makes sense to have both an isOpen and isLoaded property, the former means “is shown somewhere in the UI” and would make it appear (e.g. in file browser) with close button etc., whereas the latter means that data has been loaded from disk, and we can safely access the buffer.
It looks better when we have a “full page” of results on the first update, rather than building up the first page of results file-by-file with only 0.03 seconds between each “increment”.
