SWI-Prolog -- library(unicode): Unicode string handling

1 library(unicode): Unicode string handling

This library wraps the utf8proc library, giving Prolog code access to Unicode character properties, string normalization, case folding, and grapheme-cluster iteration.

Three levels of API are provided:

Normalization: unicode_nfd/2, unicode_nfc/2, unicode_nfkd/2, unicode_nfkc/2 implement the four standard Unicode normalization forms (NFD, NFC, NFKD, NFKC; see UAX#15). unicode_nfkc_casefold/2 combines NFKC with case folding for caseless identifier matching (see UAX#31).
Per-codepoint properties: unicode_property/2 queries the Unicode property database (general category, bidi class, decomposition type, display width, case mappings, grapheme boundary class, ...).

Mixed string-level transformations: unicode_map/3 is the workhorse; it accepts a list of flags chosen from a fixed set and performs the corresponding composition of decompose / compose / strip / lump / case-fold / grapheme-boundary-mark operations in a single pass. unicode_casefold/2 is a convenience wrapper.

Grapheme clusters (user-perceived characters) can be iterated with atom_graphemes/2 and string_graphemes/2.

Loading this library also installs a Unicode NFC normalisation hook into the SWI-Prolog kernel. The kernel's unicode_atoms policy (Prolog flag, stream property and read_term/2,3 option) uses this hook for its nfc and error modes; without the library loaded those modes raise existence_error(hook, unicode_normalize). The kernel's quoted-write rule that force-quotes atoms containing combining marks is independent and works even without this library loaded.

Lump handling:

U+0020      <-- all space characters (general category Zs)
U+0027  '   <-- left/right single quotation mark U+2018..2019,
                modifier letter apostrophe U+02BC,
                modifier letter vertical line U+02C8
U+002D  -   <-- all dash characters (general category Pd),
                minus U+2212
U+002F  /   <-- fraction slash U+2044,
                division slash U+2215
U+003A  :   <-- ratio U+2236
U+003C  <   <-- single left-pointing angle quotation mark U+2039,
                left-pointing angle bracket U+2329,
                left angle bracket U+3008
U+003E  >   <-- single right-pointing angle quotation mark U+203A,
                right-pointing angle bracket U+232A,
                right angle bracket U+3009
U+005C  \   <-- set minus U+2216
U+005E  ^   <-- modifier letter up arrowhead U+02C4,
                modifier letter circumflex accent U+02C6,
                caret U+2038,
                up arrowhead U+2303
U+005F  _   <-- all connector characters (general category Pc),
                modifier letter low macron U+02CD
U+0060  `   <-- modifier letter grave accent U+02CB
U+007C  |   <-- divides U+2223
U+007E  ~   <-- tilde operator U+223C

@see http://www.unicode.org/reports/tr15/ (UAX#15 Normalization) @see http://www.unicode.org/reports/tr29/ (UAX#29 Grapheme Clusters) @see http://www.unicode.org/reports/tr31/ (UAX#31 Identifiers) @see https://github.com/JuliaStrings/utf8proc

[det]unicode_map(+In, -Out, +Options)

Perform a Unicode mapping on In, returning Out. Options is a list that may contain any combination of the flags below; a call is roughly equivalent to utf8proc_map(In, Options) in the C API.

stable: Respect Unicode versioning stability --- the result does not depend on which (recent) version of Unicode is in use.
compat: Use compatibility decomposition (i.e. formatting information is lost).
compose: Produce a composed result (e.g. NFC or NFKC, depending on the presence of compat).
decompose: Produce a decomposed result (NFD/NFKD).
ignore: Strip "default ignorable" characters (e.g. soft hyphen, zero-width space).
rejectna: Raise an error instead of returning output when the input contains unassigned code points.
nlf2ls: Convert all NLF-sequences (LF, CRLF, CR, NEL) to U+2028 LINE SEPARATOR.
nlf2ps: Convert all NLF-sequences to U+2029 PARAGRAPH SEPARATOR.
nlf2lf: Convert all NLF-sequences to U+000A LINE FEED.
stripcc: Strip or convert control characters. NLF-sequences become a space, except if one of the NLF-conversion flags is set; HT and FF are treated as NLF in this case. All other control characters are removed.
casefold: Apply Unicode case folding (for caseless comparison).
charbound: Insert a U+00FF byte at the beginning of every grapheme cluster (UAX#29). The result can be split on 0xFF to recover individual graphemes; atom_graphemes/2 wraps this pattern.
lump: Normalise typographic variants to their ASCII equivalents (see module header for the full list). Combined with nlf2lf, paragraph and line separators become U+000A as well.
stripmark: Strip all combining marks (non-spacing, spacing, enclosing). Must be combined with compose or decompose.

[det]unicode_nfd(+In, -Out)

Characters in In are decomposed by canonical equivalence (NFD). Precomposed characters expand into base + combining marks. For example U+00C5 (LATIN CAPITAL LETTER A WITH RING ABOVE) becomes the two-code sequence A + U+030A.

See also: http://www.unicode.org/reports/tr15/

[det]unicode_nfc(+In, -Out)

Characters in In are decomposed and then recomposed by canonical equivalence (NFC). Precomposed code points are preferred; for example A + U+030A becomes the single code point U+00C5.

See also: http://en.wikipedia.org/wiki/Unicode_equivalence\#Normal_forms

[det]unicode_nfkd(+In, -Out)

Characters in In are decomposed by compatibility equivalence (NFKD). Compatibility decomposition expands presentation forms (ligatures, subscripts, fullwidth letters) into their base forms; for example U+FB03 (LATIN SMALL LIGATURE FFI) becomes‘f f i`.

[det]unicode_nfkc(+In, -Out)

Characters in In are decomposed by compatibility equivalence, then recomposed by canonical equivalence (NFKC).

[det]unicode_nfkc_casefold(+In, -Out)

Equivalent to unicode_nfkc/2 followed by unicode_casefold/2 done in a single pass. This is the normalisation form recommended by UAX#31 for caseless identifier matching. For example German 'Strasse' written with U+00DF (LATIN SMALL LETTER SHARP S) maps to 'strasse', and U+FB03 (LATIN SMALL LIGATURE FFI) maps to 'ffi'.

[det]unicode_casefold(+In, -Out)

Out is the case-folded form of In. Use this for caseless comparison that does not require NFKC; otherwise prefer unicode_nfkc_casefold/2.

[nondet]unicode_property(?Code, ?Property)

Query the Unicode character database for Code. Code is an integer code point (0 .. 0x10FFFF) or a single-character atom; Property is a term of the form Name(Value) drawn from the list below.

This predicate is a thin wrapper over utf8proc's property struct, so its vocabulary matches the utf8proc documentation. In the modes (+,?) and (-,?) the predicate enumerates properties for the given code (or the code for the given property); in (+,+) it is a deterministic test.

Supported properties:

category(Atom)

Unicode general category. Atom is one of Cc, Cf, Cn, Co, Cs, Ll, Lm, Lo, Lt, Lu, Mc, Me, Mn, Nd, Nl, No, Pc, Pd, Pe, Pf, Pi, Po, Ps, Sc, Sk, Sm, So, Zl, Zp, Zs. When querying, the single capital letter of a subcategory stands for all its subcategories; e.g.

?- unicode_property(0'A, category('L')).
true.

combining_class(Integer)

Canonical combining class (0 for base characters, 230 for accents above, etc.).

bidi_class(Atom)

Bidirectional class. One of l, lre, lro, r, al, rle, rlo, pdf, en, es, et, an, cs, nsm, bn, b, s, ws, on.

bidi_mirrored(Bool)

true if the character is mirrored for bidi (parentheses, brackets, math operators, ...).

decomp_type(Atom)

Compatibility decomposition type. One of font, nobreak, initial, medial, final, isolated, circle, super, sub, vertical, wide, narrow, small, square, fraction, compat. Fails when there is no decomposition.

ignorable(Bool)

true if the character is a "default ignorable" code point.

boundclass(Atom)

UAX#29 grapheme-cluster break class. One of start, other, cr, lf, control, extend, l, v, t, lv, lvt, regional_indicator, spacingmark, prepend, zwj, extended_pictographic, e_zwg.

width(Integer)

Display width in fixed-width cells, 0..3. Zero for combining marks and control characters, 1 for most, 2 for "wide" characters (CJK, emoji).

ambiguous_width(Bool)

true if the character has East-Asian Ambiguous width --- normally one column, but two in a legacy CJK context.

uppercase(Code)

lowercase(Code)

titlecase(Code)

Single-code-point case mapping. Fails when the code point has no mapping of that kind (e.g. unicode_property(0'A, uppercase(_)) fails because 'A' is already upper-case). For characters whose case mapping produces more than one code point (e.g. U+00DF LATIN SMALL LETTER SHARP S maps to "SS"), use unicode_map/3 with the [casefold] option or unicode_casefold/2 for a full string-level transformation.

indic_conjunct_break(Atom)

Indic_Conjunct_Break property (Unicode 15+; UAX#44). One of none, linker, consonant, extend. Used by the grapheme-cluster-break algorithm for Devanagari, Bengali, etc.

See also: http://www.unicode.org/reports/tr44/ (Unicode property database)

[det]atom_graphemes(?Atom, ?Graphemes)

Relate Atom to a list of its grapheme clusters. Grapheme clusters are "user-perceived characters" as defined by UAX#29 --- e.g. the precomposed U+00E9 (LATIN SMALL LETTER E WITH ACUTE) and the decomposed sequence e + U+0301 are both one grapheme, an emoji ZWJ sequence such as MAN + ZWJ + WOMAN + ZWJ + GIRL is one grapheme, and a regional-indicator pair (e.g. U+1F1F3 U+1F1F1, rendered as the Dutch flag) is one grapheme.

In the forward mode (+Atom, ?Graphemes), Atom is decomposed into a list of atoms, each covering one cluster. In the reverse mode (?Atom, +Graphemes), the elements of Graphemes are concatenated into Atom. Both arguments instantiated means both modes run and the result must agree.

?- atom_codes(A, [0'c, 0'a, 0'f, 0'e, 0x0301]),
   atom_graphemes(A, Gs).
Gs = [c, a, f, G],
atom_codes(G, [0'e, 0x0301]).

?- atom_graphemes(A, [a, b, c]).
A = abc.

See also: string_graphemes/2 for the string analogue.

[det]string_graphemes(?String, ?Graphemes)

As atom_graphemes/2, but the elements of Graphemes are strings.

[det]unicode_version(-Version)

Version is an atom describing the Unicode version implemented by the linked utf8proc library, e.g. '15.1.0'. This drives the normalisation, case-folding and grapheme-cluster predicates in this module, and may differ from the Unicode version of the SWI-Prolog source syntax classifier reported by the read-only Prolog flag unicode_syntax_version.

[semidet]unicode_codepoint_valid(+Code)

True when Code is a non-negative integer that is a valid and assigned Unicode code point. Unassigned code points (general category Cn), surrogate halves (Cs), and integers outside‘0..0x10FFFF` all fail.