Fix CLDR pattern letters left in locale formats

[kamthorn]

Fixes #3371

Problem

Some locale formats were copied from CLDR patterns without converting them to isoFormat() tokens. In CLDR, yy is a 2-digit year and d is the day of the month. In isoFormat(), yy is not a token (so it's printed literally), and d is the day of the week:

Locale	Format	Before	After
es_PH	L	30/9/yy	30/9/26
ne_IN	L	Ly/9/3 (the leading L is the macro bug fixed in #3369)	26/9/30
ps, ps_AF	L	2026/9/3	2026/9/30
seh	LL	3 de Set de 2026	30 de Set de 2026
fur, fur_IT	LL	30 3i setembar 3pm30. 9. 26 2026	30 di setembar dal 2026

(Examples use 2026-09-30, a Wednesday.)

In fur, the words "di" and "dal" weren't escaped, so d, a and l were read as tokens.

Changes

One commit per file. Each commit message cites the CLDR 48.2.2 pattern it was checked against (cldr-json, cldr-dates-full/main//ca-gregorian.json).

File	Key	Before	After	CLDR 48.2.2
es_PH	L	D/M/yy	D/M/YY	d/M/yy
fo_DK	L	DD.MM.yy	DD.MM.YY	dd.MM.yy
hr_BA	L	D. M. yy.	D. M. YY.	d. M. yy.
ms_BN	L	D/MM/yy	D/MM/YY	d/MM/yy
ms_SG	L	D/MM/yy	D/MM/YY	d/MM/yy
ne_IN	L	yy/M/d	YY/M/D	yy/M/d
nnh	L	DD/MM/yy	DD/MM/YY	dd/MM/yy
nnh	LLL, LLLL	[lyɛ]̌ʼ d [na] MMMM…	[lyɛ]̌ʼ D [na] MMMM…	'lyɛ'̌ʼ d 'na' MMMM, y
pa_Guru	L	D/M/yy	D/M/YY	d/M/yy
sr_Cyrl_BA	L	D.M.yy.	D.M.YY.	d. M. y.
sr_Latn_BA	L	D.M.yy.	D.M.YY.	d. M. y.
ta_MY	L	D/M/yy	D/M/YY	d/M/yy
ta_SG	L	D/M/yy	D/M/YY	d/M/yy
uz_Cyrl	L	DD/MM/yy	DD/MM/YY	dd/MM/yy
fur_IT	LL, LLL, LLLL	DD di MMMM dal YYYY…	DD [di] MMMM [dal] YYYY…	d 'di' MMMM 'dal' y
ps	L	YYYY/M/d	YYYY/M/D	y/M/d
seh	LL, LLL, LLLL	d [de] MMM…	D [de] MMM…	d 'de' MMM 'de' y

sr_Cyrl_XK, sr_Latn_XK (inheriting from _BA), ps_AF and fur get the fix through their parent file.

Only the token letters changed. Everything else in each format was left as it was, including zero-padding and separators. There's one exception: CLDR 48 now uses d. M. y. (4-digit year, with spaces) for sr_*_BA. I kept the 2-digit year that the current format intends, so the only change is yy → YY. I can switch them to D. M. YYYY. if you'd prefer following current CLDR.

Tests

New tests/Language/LocaleFormatsTest.php:

• testFormatsHaveNoCldrOnlyLetters goes through every locale and fails if a format (LT…llll) or calendar format has an unescaped y or N. Those are CLDR year/era letters and are printed literally today. Before this PR, it lists 15 locales.
• testCldrFormatsUseIsoTokens checks the output of each fixed format.

No existing expected value had to change. The full phpunit suite, phpstan and phpmd pass locally on PHP 8.4, and php-cs-fixer reports nothing for the changed files.

Notes

• nnh LLL/LLLL start with [, so they're also hit by the macro bug fixed in Fix first character after macro expansion in isoFormat() #3369. Their output is only fully right with both PRs, which is why the tests here only check nnh L.
• I added my name to the Authors header of the 3 files that have one (hr_BA, fur_IT, ps), following 5f0ac3b.

This change was written with the help of Claude Code (AI). I reviewed the diff myself.

🤖 Generated with Claude Code