Chrome specific issue that happens when destroying a chart and re-creating it immediately (same animation frame?). The CSS animation used to detect when the canvas become visible is not re-evaluated, breaking responsiveness. Accessing the `offsetParent` property will force a reflow and re-evaluate the CSS animation.
Inject the GA tracking snippet for all samples, including the index page. Also update README.md badges using the shields.io service for consistency with flat-square style and cache, and add release badges to the installation documentation page.
If `window` or `document` are `undefined`, a minimal platform implementation is used instead, which one only returns a context2d read from the given canvas/context.
Remove the special drawing for an undefined zero grid line since it causes issue when the first tick is not aligned on the scale extremity (ie only linear scales now display a special grid line for the origin). Hide scales in the filler plugin unit test fixtures to avoid future failures due to changes unrelated to the tested features.
New `options.resolve` helper that determines the final value to use from an array of input values (fallback) and a given context and/or index. For now, only the bubble chart support scriptable options, see documentation for details.
Add scriptable options documentation and update the bubble chart dataset properties table with their scriptable and indexable capabilities and default values. Also move point style description under the element configuration section.
Add a new `offset` option to scales to add extra space at edges and remove the `includeOffset` argument from `getPixelForValue()` and `getPixelForTick()`. The bar controller now automatically calculates the bar width to avoid overlaps. When `offsetGridLines` is true, grid lines move to the left by one half of the tick interval, and labels don't move.
For consistency, enable ESLint `one-var` rule to require multiple variable declarations for initialized variables per scope. Uninitialized variables can still be declared together (preferred) or separately.
http://eslint.org/docs/rules/one-var
The `bounds` option makes more sense directly under `scale` since it defines the scale limits strategy when no explicit min/max are specified. Also change the `bounds: 'labels'` option value for `bounds: 'ticks'` because it really means: "ensure ticks to be fully visible in the scale, whatever the ticks `source`.
Allow to create a chart on a canvas not yet attached to the DOM (detection based on CSS animations described in https://davidwalsh.name/detect-node-insertion). The resize element (IFRAME) is added only when the canvas receives a parent or when `style.display` changes from `none`. This change also allows to re-parent the canvas under a different node (the resizer element following). This is a preliminary work for the DIV based resizer.
Fix `ticks.mode` behavior when `ticks.source` is `auto`: the lookup table is now built from the data and not from the ticks, so data (and ticks) are correctly distributed along the scale. Rename the option to `distribution` (more explicit than `mode`) and since this option applies from now on the data, it seems better to have it under `scale` instead `scale.ticks`.
Internal ticks are now stored as objects in the PRIVATE this._ticks member and must not be accessed directly from outside this class. this.ticks is around for a long time and hasn't been marked as private, so we can't change its structure without unexpected breaking changes. If you need to access the scale ticks, use scale.getTicks() instead.
The original implementation tries to intercept events from the chart internal iframe, which ones failing on Chrome 60. Checking internals doesn't seem the best approach, instead we could consider that a chart has been resized after the resize method has been called and processed. So let's hook `Chart.resize` and callback once it's done.
`ticks.bounds` (`'data'`(default)|`'label'`): `data` preserves the data range while `labels` ensures that all labels are visible. This option is bypassed by the min/max time options.
Remove the useless time scale `_model` object containing private members: instead, make these members private (prefixed by `_`) part of the scale.
