#### Is this adding or improving a _feature_ or fixing a _bug_?
This should fix `Unable to get property 'childNodes' of undefined or null reference` error in IE11 when using slate-html-serializer and deserializing an empty string (#1757, #1756).
#### What's the new behavior?
The `defaultParseHtml` now returns an empty body element if the parsed body is null.
I tested this in the latest Chrome, Firefox and Safari, and they all seemed to work similarly. If the html param was an empty string, `parseFromString` returned a document element, which had a body property which was an empty body element. In IE11 though the body property was null, which caused the undefined or null reference later.
#### Have you checked that...?
* [x] The new code matches the existing patterns and styles.
* [x] The tests pass with `yarn test`.
* [x] The linter passes with `yarn lint`. (Fix errors with `yarn prettier`.)
* [x] The relevant examples still work. (Run examples with `yarn watch`.)
* Run mocha test with module alias
* Running test with babel module alias
* Fix model alias
* Fix model alias
* Resolve module alias
* Running test with babel module alias
* Connect to codecov
* add codecov to travis
* stop if yarn test has errors
* Still cannot collect data from slate modules
* Try to check whether it works with codecov
* Move config to nycrc
* Remove nyc require
* Update nyc to use src
* better before_script
#### Is this adding or improving a _feature_ or fixing a _bug_?
Feature.
#### What's the new behavior?
This pull request adds paths to `Range` objects, including the selection. The paths and keys are kept in sync automatically, so that you can use whichever is ideal for your use case.
This should allow us to use paths for lots of the internal logic, which are much quicker to work with than keys since they avoid having to lookup the key in the document and can just traverse right to the node in question.
#### How does this change work?
`Range` objects have two new properties:
```js
range.anchorPath
range.focusPath
```
(Eventually these will be `range.anchor.path` and `range.focus.path` when points are introduced.)
When operations occur and whenever ranges are created/normalized, the paths are updated and kept in sync with the keys.
#### Have you checked that...?
<!--
Please run through this checklist for your pull request:
-->
* [x] The new code matches the existing patterns and styles.
* [x] The tests pass with `yarn test`.
* [x] The linter passes with `yarn lint`. (Fix errors with `yarn prettier`.)
* [x] The relevant examples still work. (Run examples with `yarn watch`.)
#### Does this fix any issues or need any specific reviewers?
Fixes: https://github.com/ianstormtaylor/slate/issues/1408
Fixes: https://github.com/ianstormtaylor/slate/issues/1567
* Use slate rather than relative path
* Move benchmark to one dir
* Use slate-* instead of relative path
* Before and After Function
* Remove un-necessary cross-env
* Hard fix
* Lint the hard fix
* Reset memory in bench()
* Working on Benchmark Frameworks
* Rename to slate-dev-benchmark
* Add packages
* Fix prettier bug
* Benchmark framework is in working
* Do not log in test
* max times test
* mute logger in test
* add hr time
* Better support for maxTime; add support of split runs to save memory space
* Fix maxTries
* Add global.gc
* Global gc for each bench
* Better test interface
* Test max-time
* Test max-time done
* Add Benchmark among packages
* Starting to get benchmark running
* Pure Node lib
* Change babelrc for pure Node benchmark
* Moving Benchmarks
* Get benchmark and test running
* Get benchmark for slate-html-serializer
* add slate-react
* add slate/changes
* all benchmarks are converted
* Run benchmark by yarn
* Run benchmark with expose-gc
* Annotate Bench.js
* Do not bundle slate-dev-benchmark in rollup
* Add annotation
* Allow config file to enable part benchmark compare
* Add config for compare
* support compare.js
* Do not re-allocate memory; due to a large heap taken influence result
* Render with Decorations
* get active marks at range
* Fix bug in showing percents
* Fix percent showing bug
* chore: add more benches
* Better output of benchmark
* Fix linting
* decoration and normal as different benchmark test
* Fix deserialize benchmark
* README.md
* Fix Readme.md
* README.md
* block-spacing config
* safer user config loading
* use package.json to load package in test
* Consistent linting
* move components to parent directory
* Annotation styling in package
* margin line before multi-line block
* Fix naive bug
* Fix naive bug
* Fix a blank line
* only log user and hr
* Better name
* Better annotation for runBundleTasks
* Fix typo
* Better logger
* Move async to test
* Omit skip
* Only log the user space time
* Single line async sleep
* file name fix
* Fix annotation
* Better output of compare
* Remove get-characters(-at-range) benchmarks
* Restore emoji
* Capitalize types
* Remove compare to another area
* Add grep and config interface
* Linting files
* Linting benchmarks
* Linting benchmarks
* Update yarn.lock
Sometimes, I have nodes that only make sense temporarily, so it
doesn't make sense to serialize them. Following the pattern in React,
explicitly returning `null` from a serialzation rule should result in
that node (and its children) not making it into the serialized document.
