Checks the first recognised locale wins over later and earlier unrecognised
ones, malformed entries are skipped, and an unrecognised or missing locale
returns English (United States) with exactly one notice that names the first
detected locale or says the host reports none.
ResolveHostLanguage takes the host locales in order of preference from a source
and the console language of the first one that is recognised. When none is
(or the source returns none), it reports one notice line that names the
detected locale and the English (United States) fallback and returns it, so
auto never fails. Source and notice are injectable so tests do not depend on
the host.
The macOS locale source calls CFLocaleCopyPreferredLanguages. The header is
inline, so every target that resolves the language needs the framework: an
interface target carries it on macOS and nothing elsewhere, and the system
service library, the trophy library and the locale test link it.
MacHostLocales returns the user's preferred languages in order from
CFLocaleCopyPreferredLanguages, BCP-47 tags such as en-US, zh-Hans-CN or
pt-BR, which the process environment of a macOS desktop session does not
carry. Only built on macOS.
WindowsHostLocales returns the user default locale name from
GetUserDefaultLocaleName, a BCP-47 tag such as en-US, zh-Hans-CN or pt-BR. It
reads the regional-format locale of the user, not the list of preferred
display languages, because that is one documented call with a fixed buffer.
A name with a non-ASCII character is treated as no locale. Only built on
Windows.
Checks the LC_ALL, LC_MESSAGES, LANG precedence with empty and unset
variables, that other categories are not read, and the LANGUAGE rules: its
entries come first under a real locale, empty entries are skipped, and it is
ignored under C, POSIX, C.UTF-8 and when nothing else is set.
PosixHostLocales takes an injectable environment reader and returns the locales
in order of preference: LC_ALL, then LC_MESSAGES, then LANG, the first one that
is set and not empty, as the C library chooses the locale of the messages
category. When that locale is a real one (not C, POSIX or unset) the entries of
LANGUAGE, the GNU preference list, come before it, as gettext ignores LANGUAGE
under the C locale.
Checks that every console language is reached by a tag of its own, that
unrecognised locales (empty, C, POSIX, malformed, languages the console has no
entry for such as lt, he, hi, ca and sr) give English (United States) and are
reported as unrecognised, and that a recognised language with an unknown
region or script takes its main row.
MatchLocale returns the console language of a locale and whether the locale was
recognised. An empty, C, POSIX, malformed or unlisted locale is not
recognised and gives English (United States), the language the console answers
when ANYPS5_LANGUAGE is unset, so resolving a host locale cannot fail. A
recognised language with no row for its region already takes the language's
main row.
Each has a single console language, so every region maps to it. The legacy
code in for Indonesian, still used by some POSIX and Java hosts, is accepted
as well as id.
Checks every language of the group, the three Norwegian codes, and that the
Baltic and other Slavic languages the console has no entry for (lt, lv, et, sk,
sl, bg, sr, hr, be, is) are not mapped to a neighbour.
Russian, Ukrainian, Polish, Czech, Finnish, Swedish and Danish map to their
console language. Norwegian is the console's one Norwegian language, so no,
nb (Bokmal) and nn (Nynorsk) all map to it. Lithuanian, Latvian, Estonian and
the other Slavic languages the console has no entry for stay unrecognised.
Checks Simplified and Traditional by region (CN, SG, TW, HK, MO), by script
(Hans, Hant), a script that contradicts the region (zh-Hans-TW, zh-Hant-CN),
an unknown script and region falling back to Simplified, and the Mandarin and
Cantonese rows.
A script decides first (zh-Hant is Traditional, zh-Hans Simplified, also
against a contradicting region such as zh-Hans-TW), then the region (TW, HK and
MO are Traditional), and a bare zh is Simplified. Mandarin (cmn) follows the
same rules and Cantonese (yue) is Traditional unless it names Hans.
Checks the Spain row, every Latin American country row, the es-419 region, the
Brazilian default for a bare pt, the Portuguese territories, and spellings
with case differences, POSIX codesets and an explicit script.
The console has Spanish (Spain) and Spanish (Latin America), and Portuguese
(Portugal) and Portuguese (Brazil). A bare es is Spain, and the Latin American
countries, the es-419 region and es-US are Latin America. A bare pt is Brazil,
the language's larger variant, and Portugal and the African and Asian
territories that follow the European standard (AO, MZ, CV, GW, ST, TL, MO) are
Portugal.
Maps English, French, German, Italian, Dutch and Romanian locales in BCP-47
and POSIX spellings, including the regional rows (en-GB, en-IE, en-AU, en-NZ,
fr-CA) and the regions that fall back to the main row (en-CA, en-IN, fr-BE,
fr-CH, de-AT, de-CH, nl-BE).
ConsoleLanguageForTag picks, among the rows of the tag's language whose script
and region are empty or equal to the tag's, the one that names the most
(script outranks region), so a regional row wins over the language's main row
and a language with no row for the tag's region falls back to its main row.
The first group of rows covers the Latin-script European languages that have
no regional split but a few: English (United Kingdom for GB, IE, AU and NZ,
English (United States) otherwise), French and French (Canada), German,
Italian, Dutch and Romanian.
Checks that BCP-47 and POSIX spellings of the same locale give the same
language, script and region, that the @latin modifier supplies a script, and
that C, POSIX, empty and malformed texts are refused without writing the tag.
ParseLocale normalises the text as a POSIX name first, which leaves a BCP-47
tag unchanged except for underscores, and then parses the result, so the
caller handles en-US and en_US.UTF-8 alike. C, POSIX and an empty text report
no locale and leave the output tag as it was.
Checks NormalizePosixLocale on codesets, modifiers, the Latn and Cyrl script
modifiers (also where the name already has a script), names that are already
BCP-47, and C, POSIX and empty names, which must normalise to no locale.
POSIX hosts name their locale language[_territory][.codeset][@modifier], such
as en_GB.UTF-8@euro, and use C and POSIX for no locale at all. NormalizePosixLocale
drops the codeset and the modifier, turns underscores into hyphens, maps the
@latin and @cyrillic modifiers to the Latn and Cyrl scripts where the name has
no script, and returns an empty string for C, POSIX and an empty name so the
caller sees "no locale" instead of a language called C.
Adds a host-only test, system_service_language_locale, for ParseTag: case
normalisation of the language, script and region subtags, the three-digit
region of es-419, variants and extensions that are ignored, and the inputs it
must reject (empty and oversized subtags, one-letter and long languages, the
POSIX underscore and codeset forms) without touching the output tag.
ANYPS5_LANGUAGE can only name a console language number. Taking the language
from the host needs the host's locale, which arrives as a BCP-47 tag such as
en-US, pt-BR or zh-Hant-TW. ParseTag reads the language, script and region
subtags, normalises their case, ignores variants and extensions, and rejects
empty subtags, subtags over 8 characters and languages that are not 2 or 3
letters.
MarcZX10