[refactor] Refactoring of hyperscript.js and render.js, including performance improvements

[kfule]

Refactor hyperscript.js and render.js. In particular, the replacement of fix #2622 appears to have significantly improved the performance regression.

Description

This change is a refactoring aimed at improving code outlook and performance.

• hyperscript.js now uses Object.assign() to merge attrs. Also, Object.assign() controls the order in which attributes are applied, which addresses issue Error when using CSS selector for input type and passing a special attribute #2622 alternatively.
• In render.js, performance was improved by eliminating the past Error when using CSS selector for input type and passing a special attribute #2622 fix. Also, isFileInput has been moved into setAttr() to improve code outlook.
• Polyfill for Object.assign() is removed. The polyfill doesn't support multiple source objects (like in this pr), and almost all browsers now natively support Object.assign().

results of `npm run perf`

before (v2.2.8)

construct large vnode tree x 18,987 ops/sec ±0.79% (126 runs sampled)
rerender identical vnode x 3,898,180 ops/sec ±0.52% (127 runs sampled)
rerender same tree x 55,444 ops/sec ±1.71% (128 runs sampled)
add large nested tree x 6,701 ops/sec ±1.22% (123 runs sampled)
mutate styles/properties x 78.77 ops/sec ±3.08% (75 runs sampled)
repeated add/removal x 3,052 ops/sec ±1.15% (126 runs sampled)
Completed perf tests in 60100ms

after ("rerender same tree" is improved)

construct large vnode tree x 19,972 ops/sec ±0.81% (123 runs sampled)
rerender identical vnode x 3,889,751 ops/sec ±1.03% (127 runs sampled)
rerender same tree x 88,338 ops/sec ±0.27% (125 runs sampled)
add large nested tree x 7,014 ops/sec ±1.61% (119 runs sampled)
mutate styles/properties x 77.55 ops/sec ±1.12% (73 runs sampled)
repeated add/removal x 3,067 ops/sec ±2.48% (90 runs sampled)
Completed perf tests in 59057ms

cf. v2.0.4 (by backported test-perf.js and m.censor)

construct large vnode tree x 13,149 ops/sec ±0.51% (124 runs sampled)
rerender identical vnode x 5,084,840 ops/sec ±0.49% (127 runs sampled)
rerender same tree x 82,781 ops/sec ±0.18% (128 runs sampled)
add large nested tree x 6,611 ops/sec ±2.56% (101 runs sampled)
mutate styles/properties x 79.76 ops/sec ±0.86% (73 runs sampled)
repeated add/removal x 3,067 ops/sec ±0.97% (125 runs sampled)
Completed perf tests in 59579ms

As shown above, npm run perf is improved, especially for "rerender same tree".

I have also run test-perf.js in a real browser to see the performance improvement.
(If necessary, I would like to measure and tabulate performance later.)

Motivation and Context

As for performance, it is sufficiently sophisticated at v2.0.4. However, subsequent bug fixes and enhancements seem to have caused some performance regression. I measured the performance of v2.0.4 and later versions with npm run perf and found that 7c7d76d of #2578 has a negative impact on performance.
The commit 7c7d76d fixes #2622 by changing setAttr(s), but I noticed that it can be resolved by hyperscript(execSelector) instead of setAttr(s).

How Has This Been Tested?

• npm run test
• I also confirmed with flems that Error when using CSS selector for input type and passing a special attribute #2622 is not reproduced.

Types of changes

• Bug fix (non-breaking change which fixes an issue)
• New feature (non-breaking change which adds functionality)
• Breaking change (fix or feature that would cause existing functionality to change)

Checklist

• My code follows the code style of this project.
• I have added tests to cover my changes.
• All new and existing tests passed.
• My change requires a documentation update, and I've opened a pull request to update it already:
• I have read https://mithril.js.org/contributing.html.

[dead-claudia]

One minor nit, then it's got my approval.

(This will cause some merge conflicts for my #2982, but it wouldn't be the first time I rebuilt that PR.)

[dead-claudia]

Could you just drop this module entirely and use Object.assign directly in the places that use it?

Shouldn't impact performance, but will provide a small bundle size win and it's one less source file to worry about.

[kfule]

OK, I removed the module.

[dead-claudia]

@kfule Sorry for the delayed review. Been hard at work nailing down a long math-heavy justification for one of the internal workflow changes in an upcoming update to #2982.

[kfule]

@dead-claudia
Thanks for your review.
I removed the assign.js module as you suggested.
Also, I removed unnecessary parameters from setAttr().

[dead-claudia]

Oops, meant to squash that, but was also debugging Refined GitHub at the same time. 🙃