[dylanpulver]

IANA_SUPPORTED folds in IANA_NO_ALIASES, but _IANA_NAMES is built only from encodings.aliases. Those nine names are in IANA_NO_ALIASES because they have no alias entry, so seven of them never reach _IANA_NAMES — cp720, cp737, cp856, cp875, cp1006, koi8_t, koi8_u. cp874 and koi8_r resolve only because they happen to also appear in aliases.values(), which is why this has been quiet.

koi8_r vs koi8_u is the whole bug in one pair:

koi8_r  in IANA_SUPPORTED: True   iana_name('koi8_r') -> 'koi8_r'
koi8_u  in IANA_SUPPORTED: True   iana_name('koi8_u') -> ValueError: Unable to retrieve IANA for 'koi8_u'

Three consequences, all measured on master (e239bdc):

	koi8_r	koi8_u
any_specified_encoding on a page declaring it	koi8_r	None
from_bytes(...).best().encoding for that page	koi8_r	shift_jis_2004
header after .output('utf_8')	rewritten to utf-8	left saying koi8-u
.output('koi8_u')	—	ValueError

So a declared charset is ignored, the detector then picks the wrong encoding, and the re-encoded payload keeps a <meta charset="koi8-u"> that is no longer true. #328 added these nine to make them supported and updated docs/user/support.rst accordingly; only the lookup table was left behind.

What I ran. Same script on master and on the patch: iana_name() over all of IANA_NO_ALIASES — 2/9 resolvable before, 9/9 after; the declared-charset round trip above for koi8_r/koi8_u/cp1251; and .output() to four Cyrillic encodings. nox -s test (as CONTRIBUTING asks): the test-3.14 session gives 197 passed on master and 297 passed here — the added test is parametrised over IANA_SUPPORTED, which is 100 entries. Reverting only constant.py and keeping the test fails it exactly 7 times, one per unresolvable name, so it isn't vacuous. Full nox -s test is green on 3.9–3.14 and pypy (3.7/3.8/3.15 skipped, interpreters not installed here). nox -s lint passes every hook — ruff, ruff format, mypy, cython-lint, codespell, zizmor.

The alternative I rejected, and it also passes. Building the self-map from IANA_SUPPORTED instead is a one-liner that passes every check above and the full suite. It also silently drops eight names _IANA_NAMES resolves today — rot_13, mbcs, base64_codec, bz2_codec, hex_codec, quopri_codec, uu_codec, zlib_codec — because IANA_SUPPORTED filters those out. Adding IANA_NO_ALIASES keeps the change additive.

Not addressed: CHARDET_CORRESPONDENCE keys off iso8859_1, which is not in IANA_SUPPORTED (Python's canonical name is latin_1), so that entry looks unreachable too — separate from this and left alone.

AI assistance: found and patched with Claude Code (Claude Opus 5, claude-opus-5); I reviewed and ran everything above myself.

[Ousret]

LGTM