coreui / coreui-pro / 35587196909

94%

main: 94%

push

github

perf(Calendar): format dates through cached Intl formatters (#1375)

Every cell with formatting options built its own formatter, because
`toLocaleDateString(locale, options)` constructs an `Intl.DateTimeFormat` per
call. A two-panel days render makes 116 such calls: two month names, two
years, 28 weekday headers and 84 day numbers. The other 84 calls, the
aria-labels, pass no options and the engine already caches that path, so they
were never the cost: measured in node, 84 calls with options take 3.17 ms and
84 without take 0.07 ms.

The instance keeps a Map of formatters keyed by locale and options. The locale
is part of the key, so nothing has to invalidate the map when `setConfig()`
changes it, and the pickers, which push a config on every selection, cost
nothing. A spec counts the constructions: a two-panel render builds fewer than
ten formatters and more than zero, which is what a revert would produce.

An unparseable date keeps its old output rather than taking the calendar down.
`Intl.DateTimeFormat.prototype.format` throws `RangeError: Invalid time value`
where `toLocaleDateString` returns the string `Invalid Date`, and a week
selection reaches that: a string that misses every pattern falls through to an
Invalid Date, which `_initializeDates` does not replace because it only catches
null. Measured: without the guard, `new Calendar(el, { selectionType: 'week',
calendarDate: 'next week' })` throws out of the constructor and renders
nothing.

Measured in node over the call pattern of a two-panel render: 3.76 ms to
0.60 ms. What remains is mostly `getMonthsNames()`, which builds twelve
formatters per panel in every view; that is its own change.

5812 of 6426 branches covered (90.45%)

Branch coverage included in aggregate %.

10 of 10 new or added lines in 1 file covered. (100.0%)

11004 of 11393 relevant lines covered (96.59%)

556.34 hits per line