Use NumberFormatter instead of legacy NumberFormat

[rickmacgillis]

Description

The following code:

<?PHP
function nth(int $number) : ?string {
		$formatter = new \NumberFormatter('en_US', \NumberFormatter::ORDINAL);
		$formatted = $formatter->format($number);
		return $formatted === false ? null : $formatted;
}

echo nth(651959195189);
?>

Resulted in this output:

651,959,195,189rd

But I expected this output instead:

651,959,195,189th

PHP Version

PHP 8.3.6

Operating System

Linux Mint 22

[damianwadley]

Pretty confident this is a bug in ICU, but I'll let someone else make that call.

What seems to be happening behind the scenes is that ICU is taking the 64-bit integer given to it by PHP and converting it into a 32-bit integer (I assume accidentally) getting -875833803 as a result, looking up the ordinal formatting rules (which are essentially switch (number % 10) { ... }), and formatting the original 64-bit number using the rule for ...3 (which is of course {number}rd).

Good news is that there's a workaround: if you force the number to a float, that means it will get passed to ICU's unum_formatDouble which appears to work correctly.
https://3v4l.org/C0WgV

[rickmacgillis]

Confirmed. Type hinting float instead of int fixed the issue.

[cmb69]

Indeed, looks like an upstream issue:

https://github.com/unicode-org/icu/blob/e025466e3ac94bcf4a8062b98b68f4c6b7e00990/icu4c/source/i18n/nfrule.cpp#L775-L776

But we should also read https://unicode-org.github.io/icu/userguide/format_parse/numbers/legacy-numberformat.html:

⚠️ Since ICU 60, the recommended way to format numbers is NumberFormatter; see index.md.

This page is here for reference for the older NumberFormat hierarchy in ICU4C and ICU4J.

[cmb69]

Okay, no further feedback and apparently an upstream issue, so I'm repurposing this ticket.