Languages

Every language the package supplies, what its members are, what decides membership, and where a guide shows it in use. A string language is decided by Python's own reading (a character test, or the standard library's parser), never a regular expression standing in for a parser. Every language visits its hazards first: the empty string and whitespace, control and format characters, lone surrogates, combining marks, characters that change length under case mapping or normalisation, homoglyphs, non-standard spaces, overlong inputs and text that spells a missing value, plus the hazards particular to it. Records and trees are languages too, named by the schema that describes them; they are on Records and schemas and Trees.

Alphabets

Every string over a set of characters, the empty string included; L[ascii] \ {""} leaves it out.

Language	Members	Decided by	Guide
L[unicode]	every str, lone surrogates included	nothing to decide	Text
L[ascii]	the 128 ASCII code points	ord(c) < 128	Text
L[latin-1]	what the latin-1 codec encodes	ord(c) < 256	Text
L[printable]	what Python calls printable	str.isprintable	Text
L[digit]	the ten ASCII digits	c in "0123456789"	Formats
L[alpha]	letters in any script	str.isalpha	Formats
L[alnum]	letters and digits in any script	str.isalnum	Formats

Hazard alphabets

The characters parsers tend to get wrong, each a language of its own so a claim can quantify over exactly them.

Language	Members	Why it matters
L[control]	the C0 control characters and the space	newlines and NULs that end a header or a record
L[invisible]	zero-width, byte-order and bidirectional marks	text that shows nothing or reorders itself
L[combining]	combining marks	accents stacked on a letter, more code points than a reader sees
L[surrogate]	lone surrogates	a str Python holds and no encoding can write
L[compatibility]	characters NFKC maps to something else	ligatures, fullwidth letters, ℀
L[astral]	characters past the basic multilingual plane	emoji, two UTF-16 units each

Formats

Exactly what the standard library's parser accepts on the running Python.

Language	Members	Decided by	Guide
L[identifier]	Python identifiers	str.isidentifier	Formats
L[json]	what json.loads accepts, the bare NaN included	json.loads	JSON
L[uuid]	every spelling uuid.UUID reads	uuid.UUID	Formats
L[iso_date]	ISO 8601 dates	date.fromisoformat	Formats
L[iso_datetime]	ISO 8601 date-times	datetime.fromisoformat	Formats
L[ipv4], L[ipv6]	IP addresses	ipaddress	Formats
L[base64]	canonical base64	b64decode(validate=True), re-encoded equal	Formats
L[hex]	what bytes.fromhex reads	bytes.fromhex	Formats
L[slug]	lower-case words joined by single hyphens	[a-z0-9]+(-[a-z0-9]+)*	Formats
L[shell_safe]	a word the shell reads literally	shlex.quote(s) == s	Formats

L[iso_date] and L[iso_datetime] follow the running Python, which reads more ISO 8601 forms from 3.11 on, so the same claim can cover more spellings there.

Combining languages

A domain can join languages and finite sets, and leave values out:

for s in L[ascii] \ {""}, ...            every non-empty ASCII string
for s in L[slug] | {"-"}, ...            a slug, or the placeholder "-"
for s in L[unicode, len <= 80], ...      at most eighty code points

A language of your own is named the same way, by its dotted path: Writing a language.