This change also ensures that potential object-styles (user-provided or
placeholderTextColor) are group together at the end of the style array.
Fix#560
SSR was not working correctly. Styles would accumulate in the style
cache and the styles returned for each 'getApplication' call would not
represent the styles needed to render a given tree, but rather all the
styles needed to render every tree that has been rendered by that server
instance.
This is now fixed by reinitializing the style resolver after a call to
'getStyleSheet' on the server. The return type of
'AppRegistry.getApplication' is changed – { element, getStyleElement }.
The 'getStyleElement' function must be called after the component tree
has been rendered. Note that if 'getStyleElement' is not called for a
response, then its styles may leak into the next response's styles (but
will not affect the UX).
This patch also removes the 'StyleSheet.getStyleSheets' (web-only) API
and requires SSR to be done using 'AppRegistry.getApplication'.
Fix#778
Resolving styles at the same time as they are registered is problematic.
This work doesn't need to be complete until the moment a component is
rendered, rather than being done at startup.
The style resolver is reorganized to remove React Native style
registration, and contain all the stateful resolver logic (previously
partially found in NativeMethodsMixin).
Style registration is done in 'StyleSheet.create', and will subsequently
be decoupled from eager style resolution.
The stateful resolver is fully contained in the
ReactNativeStyleResolver. The options are removed from the resolver as
they aren't needed to implement the correct i18n-ified styles. This
functionality is also covered by unit tests now.
pointerEvents styles are now always resolved to classNames, even if
working with dynamic styles. In practice, inline-style pointerEvents
have always resolved to classNames as an accident of 'createElement'
registering each pointerEvent style. However, this patch ensures that
the style resolver produces correct output, and introduces the basic
mechanism by which 'placeholderTextColor' (for which classNames cannot
be precomputed) could function similarly.
WebStyleSheet handles injecting CSS rules and returning the style
sheet's text content.
All CSS, including the reset, is now added using WebStyleSheet and the DOM API 'insertRule'.
Browsers will throw a parse error on unrecognized vendor-prefixes in
selectors, so rules with unsafe selectors are wrapped in '@media all {}'
before being passed to 'insertRule'. (This will be relevant when
attempting to support 'placeholderTextColor'. And it is a way to
batch-insert rules with a single call to 'insertRule'.)
The html tag 'font-family' CSS reset is removed as it's not needed.
This change to 'benchmarks' reports the time taken to perform a forced
layout after mounting the tree. Adding a forced layout to the stress
tests can surface how different approaches to styling may affect browser
render timings.
The total time displayed is now the sum of "scripting time" (previously
total time) and "layout time". The layout time is a reflection of the
time the browser takes to perform a style recalculation and relayout of
the document.
The Benchmark component now has a 'forceLayout' prop. When it is 'true'
a forced layout is triggered on componentDidUpdate. The time taken is
added to the sample's timing data.
Reorganizes and rewrites the benchmarks. Each implementation is now
self-contained and the benchmarks can be run using a GUI. The benchmarks
themselves have been changed so that individual tests render over a
shorter time frame and more samples are taken.
