Break text into lines, graphemes, words and sentences as Unicode's tests require

UAX #14 line breaking (LB1-LB31, numbers tailored as Unicode's tests
are) and UAX #29's grapheme cluster, word and sentence boundaries, each
from generated property tables, the rules taken in order, in linear time.
They pass all of Unicode 15.1's LineBreakTest (10,274 cases),
GraphemeBreakTest (1,187), WordBreakTest (1,826) and SentenceBreakTest
(512).

1 files changed, 173 insertions, 0 deletions

diff --git a/text/linebreak.c b/text/linebreak.c
new file mode 100644
index 0000000..241e521
--- /dev/null
+++ b/text/linebreak.c
@@ -0,0 +1,173 @@
+/* Line breaking (UAX #14, Unicode 15.1): text/text.h. Each boundary is
+ * decided by the first rule, in the standard's order, that matches it; the
+ * rules that look back over spaces, marks or numbers read what a pass
+ * before them worked out for every position. */
+#include "text.h"
+#include <stdlib.h>
+
+enum { BK = VEGA_LB_BK, CR = VEGA_LB_CR, LF = VEGA_LB_LF, NL = VEGA_LB_NL, SP = VEGA_LB_SP, ZW = VEGA_LB_ZW,
+       ZWJ = VEGA_LB_ZWJ, CM = VEGA_LB_CM, WJ = VEGA_LB_WJ, GL = VEGA_LB_GL, BA = VEGA_LB_BA, BB = VEGA_LB_BB,
+       B2 = VEGA_LB_B2, HY = VEGA_LB_HY, CB = VEGA_LB_CB, CL = VEGA_LB_CL, CP = VEGA_LB_CP, EX = VEGA_LB_EX,
+       IN = VEGA_LB_IN, NS = VEGA_LB_NS, OP = VEGA_LB_OP, QU = VEGA_LB_QU, IS = VEGA_LB_IS, NU = VEGA_LB_NU,
+       PO = VEGA_LB_PO, PR = VEGA_LB_PR, SY = VEGA_LB_SY, AK = VEGA_LB_AK, AL = VEGA_LB_AL, AP = VEGA_LB_AP,
+       AS = VEGA_LB_AS, EB = VEGA_LB_EB, EM = VEGA_LB_EM, H2 = VEGA_LB_H2, H3 = VEGA_LB_H3, HL = VEGA_LB_HL,
+       ID = VEGA_LB_ID, JL = VEGA_LB_JL, JV = VEGA_LB_JV, JT = VEGA_LB_JT, RI = VEGA_LB_RI, VF = VEGA_LB_VF,
+       VI = VEGA_LB_VI };
+enum { NONE = VEGA_BREAK_NONE, ALLOWED = VEGA_BREAK_ALLOWED, MANDATORY = VEGA_BREAK_MANDATORY };
+#define NO_UNIT SIZE_MAX
+
+/* LB1: ambiguous, surrogate and unknown are alphabetic; complex context is
+ * a mark if it is one, else alphabetic; conditional Japanese starters are
+ * nonstarters. */
+static uint8_t resolve(uint32_t code) {
+    VegaLineBreak c = vega_line_break(code);
+    switch (c) {
+    case VEGA_LB_AI: case VEGA_LB_SG: case VEGA_LB_XX: return AL;
+    case VEGA_LB_SA: {
+        VegaCategory g = vega_category(code);
+        return g == VEGA_MN || g == VEGA_MC ? CM : AL;
+    }
+    case VEGA_LB_CJ: return NS;
+    default: return (uint8_t)c;
+    }
+}
+
+typedef struct {
+    const uint32_t *codes;
+    size_t count;
+    uint8_t *cls;     /* each character's class (LB1) */
+    uint8_t *eff;     /* its unit's class: a mark takes its base's (LB9), a mark with none is alphabetic (LB10) */
+    uint8_t *joined;  /* a mark that LB9 joins to the character before */
+    uint8_t *number;  /* its unit ends NU (NU | SY | IS)* */
+    size_t *start;    /* where its unit starts */
+    size_t *nonspace; /* where the last unit up to it that is not a space starts, or NO_UNIT */
+    size_t *ri;       /* how many regional indicators end at its unit */
+} Text;
+
+static bool one_of(unsigned c, unsigned a, unsigned b) { return c == a || c == b; }
+
+/* The unit after the one at i, or count. */
+static size_t next_unit(const Text *t, size_t i) {
+    size_t j = i + 1;
+    while (j < t->count && t->joined[j]) j++;
+    return j;
+}
+
+static bool aksara_start(const Text *t, size_t unit) {
+    unsigned c = t->eff[unit];
+    return c == AK || c == AS || t->codes[unit] == 0x25CC;
+}
+
+static uint8_t decide(const Text *t, size_t i) {
+    unsigned a = t->cls[i - 1], b = t->cls[i], A = t->eff[i - 1], B = t->eff[i];
+    size_t unit = t->start[i - 1], before = t->nonspace[i - 1];
+    uint32_t base = t->codes[unit];
+    if (a == BK) return MANDATORY;                                         /* LB4 */
+    if (a == CR && b == LF) return NONE;                                   /* LB5 */
+    if (a == CR || a == LF || a == NL) return MANDATORY;
+    if (b == BK || b == CR || b == LF || b == NL) return NONE;             /* LB6 */
+    if (b == SP || b == ZW) return NONE;                                   /* LB7 */
+    if (before != NO_UNIT && t->eff[before] == ZW) return ALLOWED;         /* LB8: ZW SP* ÷ */
+    if (a == ZWJ) return NONE;                                             /* LB8a */
+    if (t->joined[i]) return NONE;                                         /* LB9 */
+    if (A == WJ || B == WJ) return NONE;                                   /* LB11 */
+    if (A == GL) return NONE;                                              /* LB12 */
+    if (B == GL && A != SP && A != BA && A != HY) return NONE;             /* LB12a */
+    if (B == CL || B == CP || B == EX || B == IS || B == SY) return NONE;  /* LB13 */
+    if (before != NO_UNIT && t->eff[before] == OP) return NONE;            /* LB14: OP SP* × */
+    /* LB15a: an initial quotation mark after a start, opening or space,
+     * then spaces, and no break */
+    if (before != NO_UNIT && t->eff[before] == QU && vega_category(t->codes[before]) == VEGA_PI) {
+        if (!before) return NONE;
+        unsigned q = t->eff[before - 1];
+        if (q == BK || q == CR || q == LF || q == NL || q == OP || q == QU || q == GL || q == SP || q == ZW) return NONE;
+    }
+    /* LB15b: no break before a final quotation mark that ends something */
+    if (B == QU && vega_category(t->codes[i]) == VEGA_PF) {
+        size_t j = next_unit(t, i);
+        if (j == t->count) return NONE;
+        unsigned n = t->eff[j];
+        if (n == SP || n == GL || n == WJ || n == CL || n == QU || n == CP || n == EX || n == IS || n == SY ||
+            n == BK || n == CR || n == LF || n == NL || n == ZW)
+            return NONE;
+    }
+    if (B == NS && before != NO_UNIT && one_of(t->eff[before], CL, CP)) return NONE; /* LB16 */
+    if (B == B2 && before != NO_UNIT && t->eff[before] == B2) return NONE;            /* LB17 */
+    if (A == SP) return ALLOWED;                                           /* LB18 */
+    if (B == QU || A == QU) return NONE;                                   /* LB19 */
+    if (B == CB || A == CB) return ALLOWED;                                /* LB20 */
+    if (B == BA || B == HY || B == NS || A == BB) return NONE;             /* LB21 */
+    if (one_of(A, HY, BA) && unit && t->eff[unit - 1] == HL) return NONE;  /* LB21a */
+    if (A == SY && B == HL) return NONE;                                   /* LB21b */
+    if (B == IN) return NONE;                                              /* LB22 */
+    if (one_of(A, AL, HL) && B == NU) return NONE;                         /* LB23 */
+    if (A == NU && one_of(B, AL, HL)) return NONE;
+    if (A == PR && (B == ID || B == EB || B == EM)) return NONE;           /* LB23a */
+    if ((A == ID || A == EB || A == EM) && B == PO) return NONE;
+    if (one_of(A, PR, PO) && one_of(B, AL, HL)) return NONE;               /* LB24 */
+    if (one_of(A, AL, HL) && one_of(B, PR, PO)) return NONE;
+    /* LB25, as Example 7 of section 8.2 tailors it: numbers */
+    if (one_of(A, PR, PO)) {
+        if (B == NU) return NONE;
+        if (one_of(B, OP, HY)) {
+            size_t j = next_unit(t, i);
+            if (j < t->count && t->eff[j] == NU) return NONE;
+        }
+    }
+    if (one_of(A, OP, HY) && B == NU) return NONE;
+    if (t->number[i - 1] && (B == NU || B == SY || B == IS || B == CL || B == CP)) return NONE;
+    if (one_of(B, PO, PR) && (t->number[i - 1] || (one_of(A, CL, CP) && unit && t->number[unit - 1]))) return NONE;
+    if (A == JL && (B == JL || B == JV || B == H2 || B == H3)) return NONE; /* LB26 */
+    if (one_of(A, JV, H2) && one_of(B, JV, JT)) return NONE;
+    if (one_of(A, JT, H3) && B == JT) return NONE;
+    if ((A == JL || A == JV || A == JT || A == H2 || A == H3) && B == PO) return NONE; /* LB27 */
+    if (A == PR && (B == JL || B == JV || B == JT || B == H2 || B == H3)) return NONE;
+    if (one_of(A, AL, HL) && one_of(B, AL, HL)) return NONE;               /* LB28 */
+    /* LB28a: aksaras of Brahmic scripts hold together */
+    bool a_start = aksara_start(t, unit), b_start = aksara_start(t, i);
+    if (A == AP && b_start) return NONE;
+    if (a_start && one_of(B, VF, VI)) return NONE;
+    if (A == VI && unit && aksara_start(t, t->start[unit - 1]) && (B == AK || t->codes[i] == 0x25CC)) return NONE;
+    if (a_start && b_start) {
+        size_t j = next_unit(t, i);
+        if (j < t->count && t->eff[j] == VF) return NONE;
+    }
+    if (A == IS && one_of(B, AL, HL)) return NONE;                         /* LB29 */
+    if ((A == AL || A == HL || A == NU) && B == OP && !vega_east_asian_bracket(t->codes[i])) return NONE; /* LB30 */
+    if (A == CP && !vega_east_asian_bracket(base) && (B == AL || B == HL || B == NU)) return NONE;
+    if (A == RI && B == RI && t->ri[i - 1] % 2) return NONE;               /* LB30a */
+    if (B == EM && (A == EB || (vega_extended_pictographic(base) && vega_category(base) == VEGA_CN))) return NONE; /* LB30b */
+    return ALLOWED;                                                        /* LB31 */
+}
+
+bool vega_line_breaks(const uint32_t *codes, size_t count, uint8_t *breaks) {
+    breaks[0] = NONE; /* LB2 */
+    if (!count) return true;
+    uint8_t *bytes = malloc(count * 4);
+    size_t *sizes = malloc(count * 3 * sizeof(size_t));
+    if (!bytes || !sizes) {
+        free(bytes);
+        free(sizes);
+        return false;
+    }
+    Text t = {codes, count, bytes, bytes + count, bytes + 2 * count, bytes + 3 * count, sizes, sizes + count, sizes + 2 * count};
+    for (size_t i = 0; i < count; ++i) {
+        unsigned c = t.cls[i] = resolve(codes[i]);
+        bool mark = c == CM || c == ZWJ;
+        unsigned previous = i ? t.eff[i - 1] : SP;
+        /* LB9: a mark joins any unit but these; LB10: one that does not is alphabetic. */
+        t.joined[i] = mark && i && previous != BK && previous != CR && previous != LF && previous != NL &&
+                      previous != SP && previous != ZW;
+        t.eff[i] = t.joined[i] ? (uint8_t)previous : mark ? AL : (uint8_t)c;
+        t.start[i] = t.joined[i] ? t.start[i - 1] : i;
+        t.nonspace[i] = t.eff[i] != SP ? t.start[i] : i ? t.nonspace[i - 1] : NO_UNIT;
+        unsigned e = t.eff[i];
+        t.number[i] = t.joined[i] ? t.number[i - 1] : e == NU || ((e == SY || e == IS) && i && t.number[i - 1]);
+        t.ri[i] = t.joined[i] ? t.ri[i - 1] : e == RI ? (i ? t.ri[i - 1] : 0) + 1 : 0;
+    }
+    for (size_t i = 1; i < count; ++i) breaks[i] = decide(&t, i);
+    breaks[count] = MANDATORY; /* LB3 */
+    free(bytes);
+    free(sizes);
+    return true;
+}