rAF logic was introduced almost three years ago relative to this commit,
as a primary method for scheduling animation (see gh-1578 pull).
With it there was two substantial changes - one was explicitly mentioned
and the other was not.
First, if browser window was hidden aka `document.hidden === true`
it would immediately execute all scheduled animation without waiting
for time pass i.e. tick time become `0` instead of 13 ms of a default value.
Which created possibility for circular executions in case if `complete`
method executed the same animation (see gh-3434 issue).
And the second one - since then there was two ways of scheduling animation:
with `setInterval` and `requestAnimationFrame`, but there was a
difference in their execution.
In case of `setInterval` it waited default `jQuery.fx.interval` value before
actually starting the new tick, not counting the first step which wasn't
set to be executed through tick method (aka `jQuery.fx.tick`).
Whereas `requestAnimationFrame` first scheduled the call and executed
the `step` method right after that, counting the first call of
`jQuery.fx.timer`, `tick` was happening twice in one frame.
But since tests explicitly disabled rAF method i.e.
`requestAnimationFrame = null` and checking only `setInterval` logic,
since it's impossible to do it otherwise - we missed that change.
Faulty logic also was presented with `cancelAnimationFrame`, which couldn't
clear any timers since `raf` scheduler didn't define new `timerId` value.
Because that change was so subtle, apparently no user noticed it proving
that both `cancelAnimationFrame` and `clearInterval` code paths are redundant.
Since `cancelAnimationFrame` didn't work properly and rAF is and was a primary
used code path, plus the same approach is used in other popular animation libs.
Therefore those code paths were removed.
These changes also replace two different functions which schedule the animation
with one, which checks what type of logic should be used and executes it
appropriatley, but for secondary path it now uses `setTimeout` making it more
consistent with rAF path.
Since ticks are happening globally we also don't require to listen
`visibilitychange` event.
It also changes the way how first call is scheduled so execution of
animation will not happen twice in one frame.
No new tests were not introduced, since now `setTimeout` logic should be
equivalent to the rAF one, but one test was changed since now we actually
execute animation at the first tick.
Fixes gh-3434
Closes gh-3559
Notify full progress before resolving empty animations
Register animation callbacks before their ticker
Remove the right timer when immediately-done animations spawn more
Ref 9d822bc1c1
Fixes gh-3502
Fixes gh-3503
Closes gh-3496
Commit messages used to fail the style check if they contained a word
starting with "ref", like reformat or reference, in their message body.
Close gh-3437
Ref jzaefferer/commitplease#91
Node.js 0.12 is no longer supported by upstream. So far we had to keep
a workaround that would skip running ESLint there (or even load its task)
as it no longer supports Node.js <4; the same applied to Node smoke tests
as jsdom has also dropped support for those old Node versions. Those
workarounds have been removed now.
Ref 030191ae32
Closes gh-3478
1. Use the short name of the preset in the config.
2. Run ESLint first on non-minified files.
3. Explicitly specify environments in every config file (those settings cascade
which means we've been assuming a Node.js environment where we shouldn't have).
The code replacing @CODE in wrapper.js was written so that it expected
both the code and the next line to start in the first column. This commit
adjusts the regex so to get rid of that assumption and to work properly
regardless of number of lines with comments after this block.
While this is technically not necessary for our code, contributors sometimes
re-format the wrapper file in their pull requests and the error
messages they get don't tell them what's the real problem with their code.
Closes gh-3429
