* add MARKO_NO_MIGRATE env var to skip migration stage
* add EXPECT_DEPRECATED_FAILURES env var to expect all fixtures in a deprecated directory to fail
* add COMPLAIN_THROWS env var to cause deprecation warnings to be errors
* legacy: add _isWidget property from marko-widgets@6
* add back legacy method that was not deprecated in marko-widgets
* add data-widget property back to legacy widgets
* fix defineRenderer (legacy) to work with non-widgets as well
* assign id attribute to legacy dom nodes
* allow accessing the parent widget from a legacy child template
* getWidget/s only gets legacy widgets, keep init
* remove call to non-existant api
* don't kick off registration as this causes an issue with circular dependencies in legacy widgets
* add component names in debug mode
* fix keys and this.el in legacy widgets
* ensure widgetconfig object exists for legacy init
* ensure this.el exists before attaching __widget to it
* add back private __document to legacy widgets
* add deprecation warnings
* when writing a TemplateLiteral, it should not transpile unless generated
* proper escaping for template literals
* add printing test for TemplateLiteral node
* Add failing test cases for repeated at tag hydration with preserved content
* Fix owner component for include tag and use unmodified key for matching fragment markers
* closes#847 - allow discovering tags in the directory structure components/tag-name/tag-name.marko
* add docs for discovering tags with template name matching parent directory
* Updates the diffing algorithm to use an HTMLFragment node as an abstraction rather than keeping track of startNode and endNode all throughout the diffing algorithm.
* Uses the HTMLFragment for the <${dynamic}> tag and <include> tags to preserve server-rendered content for which the renderBody is not available in the browser.
* Component ids are based on the resulting parent tree (not the owner tree). This means we cannot rely on the ids in the global lookup, so component key/refs are now also stored on the component instance.
* Static node trees are now only auto assigned a key for the top-level node (instead of all nodes).
* Server comment starting markers now have the component's key serialized so the component can be attached to its owner
* Server comment markers no longer have the id on the closing marker, it is stack based.
* Normalize differences between hydration and client-rendering, so post mount the DOM looks the same regardless of whether a component was server or client rendered.
* fix matching up fragments when hydrating by taking components and normalized text nodes into account
* remove wrapping divs in test, should no longer be needed. address hydration issue with alternate fragment matching approach.
* add fragment to dom if there's a parentNode even if no startNode
* add test for mismatched hydration
* don't detach components before moving
* use fragments to preserve renderBody content
* use ___finishFragment as the incomplete fragment marker
* ensure fragments get destroyed properly and dom node key sequences don't continue from previous renders
* use parent/owner terminology in more places, component ids are now parent scoped, key/ref components are attached to their owner for both client + server render, server comment boundaries include the owner and the key in addition to the fully scoped component id, autokeyed dom nodes are attached to their parent, hydration now uses a stack: ids in ending comment nodes not needed, hydration checks to see if a component has been initialized and will attach keys directly to the owner if so
* add mutation guards for text/comment nodes, add mutation guard for input value
* make component-pages test better represent streaming hydration, fix html/head/body hydration in a better/more generic way
* add test for async rendered keyrefs
* add test for repeated mult-level transclusion
* Autokeyed elements are now stored on the parent rather than the owner. User assigned key/refs are still stored on the owner component. Because of this, user assigned keys are now prefixed (with @) to differentiate them from autokeys. This also has the benefit that assigning numeric keys can no longer conflict with the autokeys.
* add re-rendering the intermediate container to one of the new tests
