aboutsummaryrefslogtreecommitdiff
path: root/fontconfig/fc-lang/fc-lang.c
diff options
context:
space:
mode:
Diffstat (limited to 'fontconfig/fc-lang/fc-lang.c')
-rw-r--r--fontconfig/fc-lang/fc-lang.c577
1 files changed, 577 insertions, 0 deletions
diff --git a/fontconfig/fc-lang/fc-lang.c b/fontconfig/fc-lang/fc-lang.c
new file mode 100644
index 000000000..21d568d7e
--- /dev/null
+++ b/fontconfig/fc-lang/fc-lang.c
@@ -0,0 +1,577 @@
+/*
+ * fontconfig/fc-lang/fc-lang.c
+ *
+ * Copyright © 2002 Keith Packard
+ *
+ * Permission to use, copy, modify, distribute, and sell this software and its
+ * documentation for any purpose is hereby granted without fee, provided that
+ * the above copyright notice appear in all copies and that both that
+ * copyright notice and this permission notice appear in supporting
+ * documentation, and that the name of Keith Packard not be used in
+ * advertising or publicity pertaining to distribution of the software without
+ * specific, written prior permission. Keith Packard makes no
+ * representations about the suitability of this software for any purpose. It
+ * is provided "as is" without express or implied warranty.
+ *
+ * THE AUTHOR(S) DISCLAIMS ALL WARRANTIES WITH REGARD TO THIS SOFTWARE,
+ * INCLUDING ALL IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS, IN NO
+ * EVENT SHALL THE AUTHOR(S) BE LIABLE FOR ANY SPECIAL, INDIRECT OR
+ * CONSEQUENTIAL DAMAGES OR ANY DAMAGES WHATSOEVER RESULTING FROM LOSS OF USE,
+ * DATA OR PROFITS, WHETHER IN AN ACTION OF CONTRACT, NEGLIGENCE OR OTHER
+ * TORTIOUS ACTION, ARISING OUT OF OR IN CONNECTION WITH THE USE OR
+ * PERFORMANCE OF THIS SOFTWARE.
+ */
+
+#include "fccharset.c"
+#include "fcstr.c"
+#include "fcserialize.c"
+
+/*
+ * fc-lang
+ *
+ * Read a set of language orthographies and build C declarations for
+ * charsets which can then be used to identify which languages are
+ * supported by a given font. Note that this uses some utilities
+ * from the fontconfig library, so the necessary file is simply
+ * included in this compilation. A couple of extra utility
+ * functions are also needed in slightly modified form
+ */
+
+void
+FcMemAlloc (int kind, int size)
+{
+}
+
+void
+FcMemFree (int kind, int size)
+{
+}
+
+FcPrivate void
+FcCacheObjectReference (void *object)
+{
+}
+
+FcPrivate void
+FcCacheObjectDereference (void *object)
+{
+}
+
+int FcDebugVal;
+
+FcChar8 *
+FcConfigHome (void)
+{
+ return (FcChar8 *) getenv ("HOME");
+}
+
+static void
+fatal (const char *file, int lineno, const char *msg)
+{
+ if (lineno)
+ fprintf (stderr, "%s:%d: %s\n", file, lineno, msg);
+ else
+ fprintf (stderr, "%s: %s\n", file, msg);
+ exit (1);
+}
+
+static char *
+get_line (FILE *f, char *line, int *lineno)
+{
+ char *hash;
+ int end;
+ if (!fgets (line, 1024, f))
+ return 0;
+ ++(*lineno);
+ hash = strchr (line, '#');
+ if (hash)
+ *hash = '\0';
+
+ end = strlen (line);
+ while (end > 0 && isspace (line[end-1]))
+ line[--end] = '\0';
+
+ if (line[0] == '\0' || line[0] == '\n' || line[0] == '\032' || line[0] == '\r')
+ return get_line (f, line, lineno);
+ return line;
+}
+
+static char *dir = 0;
+
+static FILE *
+scanopen (char *file)
+{
+ FILE *f;
+
+ f = fopen (file, "r");
+ if (!f && dir)
+ {
+ char path[1024];
+
+ strcpy (path, dir);
+ strcat (path, "/");
+ strcat (path, file);
+ f = fopen (path, "r");
+ }
+ return f;
+}
+
+/*
+ * build a single charset from a source file
+ *
+ * The file format is quite simple, either
+ * a single hex value or a pair separated with a dash
+ *
+ * Comments begin with '#'
+ */
+
+static const FcCharSet *
+scan (FILE *f, char *file, FcCharSetFreezer *freezer)
+{
+ FcCharSet *c = 0;
+ FcCharSet *n;
+ int start, end, ucs4;
+ char line[1024];
+ int lineno = 0;
+
+ while (get_line (f, line, &lineno))
+ {
+ if (!strncmp (line, "include", 7))
+ {
+ FILE *included_f;
+ char *included_file;
+ included_file = strchr (line, ' ');
+ if (!included_file)
+ fatal (file, lineno,
+ "invalid syntax, expected: include filename");
+ while (isspace(*included_file))
+ included_file++;
+ included_f = scanopen (included_file);
+ if (!included_f)
+ fatal (included_file, 0, "can't open");
+ n = scan (included_f, included_file, freezer);
+ fclose (included_f);
+ if (!c)
+ c = FcCharSetCreate ();
+ if (!FcCharSetMerge (c, n, NULL))
+ fatal (file, lineno, "out of memory");
+ FcCharSetDestroy (n);
+ continue;
+ }
+ if (strchr (line, '-'))
+ {
+ if (sscanf (line, "%x-%x", &start, &end) != 2)
+ fatal (file, lineno, "parse error");
+ }
+ else
+ {
+ if (sscanf (line, "%x", &start) != 1)
+ fatal (file, lineno, "parse error");
+ end = start;
+ }
+ if (!c)
+ c = FcCharSetCreate ();
+ for (ucs4 = start; ucs4 <= end; ucs4++)
+ {
+ if (!FcCharSetAddChar (c, ucs4))
+ fatal (file, lineno, "out of memory");
+ }
+ }
+ n = FcCharSetFreeze (freezer, c);
+ FcCharSetDestroy (c);
+ return n;
+}
+
+/*
+ * Convert a file name into a name suitable for C declarations
+ */
+static char *
+get_name (char *file)
+{
+ char *name;
+ char *dot;
+
+ dot = strchr (file, '.');
+ if (!dot)
+ dot = file + strlen(file);
+ name = malloc (dot - file + 1);
+ strncpy (name, file, dot - file);
+ name[dot-file] = '\0';
+ return name;
+}
+
+/*
+ * Convert a C name into a language name
+ */
+static char *
+get_lang (char *name)
+{
+ char *lang = malloc (strlen (name) + 1);
+ char *l = lang;
+ char c;
+
+ while ((c = *name++))
+ {
+ if (isupper ((int) (unsigned char) c))
+ c = tolower ((int) (unsigned char) c);
+ if (c == '_')
+ c = '-';
+ if (c == ' ')
+ continue;
+ *l++ = c;
+ }
+ *l++ = '\0';
+ return lang;
+}
+
+typedef struct _Entry {
+ int id;
+ char *file;
+} Entry;
+
+static int compare (const void *a, const void *b)
+{
+ const Entry const *as = a, *bs = b;
+ return FcStrCmpIgnoreCase (as->file, bs->file);
+}
+
+#define MAX_LANG 1024
+#define MAX_LANG_SET_MAP ((MAX_LANG + 31) / 32)
+
+#define BitSet(map, i) ((map)[(entries[i].id)>>5] |= ((FcChar32) 1 << ((entries[i].id) & 0x1f)))
+#define BitGet(map, i) ((map)[(entries[i].id)>>5] >> ((entries[i].id) & 0x1f)) & 1)
+
+int
+main (int argc, char **argv)
+{
+ static Entry entries[MAX_LANG];
+ static const FcCharSet *sets[MAX_LANG];
+ static int duplicate[MAX_LANG];
+ static int country[MAX_LANG];
+ static char *names[MAX_LANG];
+ static char *langs[MAX_LANG];
+ static int off[MAX_LANG];
+ FILE *f;
+ int ncountry = 0;
+ int i = 0;
+ int nsets = 0;
+ int argi;
+ FcCharLeaf **leaves;
+ int total_leaves = 0;
+ int l, sl, tl, tn;
+ static char line[1024];
+ static FcChar32 map[MAX_LANG_SET_MAP];
+ int num_lang_set_map;
+ int setRangeStart[26];
+ int setRangeEnd[26];
+ FcChar8 setRangeChar;
+ FcCharSetFreezer *freezer;
+
+ freezer = FcCharSetFreezerCreate ();
+ if (!freezer)
+ fatal (argv[0], 0, "out of memory");
+ argi = 1;
+ while (argv[argi])
+ {
+ if (!strcmp (argv[argi], "-d"))
+ {
+ argi++;
+ dir = argv[argi++];
+ continue;
+ }
+ if (i == MAX_LANG)
+ fatal (argv[0], 0, "Too many languages");
+ entries[i].id = i;
+ entries[i].file = argv[argi++];
+ i++;
+ }
+ entries[i].file = 0;
+ qsort (entries, i, sizeof (Entry), compare);
+ i = 0;
+ while (entries[i].file)
+ {
+ f = scanopen (entries[i].file);
+ if (!f)
+ fatal (entries[i].file, 0, strerror (errno));
+ sets[i] = scan (f, entries[i].file, freezer);
+ names[i] = get_name (entries[i].file);
+ langs[i] = get_lang(names[i]);
+ if (strchr (langs[i], '-'))
+ country[ncountry++] = i;
+
+ total_leaves += sets[i]->num;
+ i++;
+ fclose (f);
+ }
+ nsets = i;
+ sets[i] = 0;
+ leaves = malloc (total_leaves * sizeof (FcCharLeaf *));
+ tl = 0;
+ /*
+ * Find unique leaves
+ */
+ for (i = 0; sets[i]; i++)
+ {
+ for (sl = 0; sl < sets[i]->num; sl++)
+ {
+ for (l = 0; l < tl; l++)
+ if (leaves[l] == FcCharSetLeaf(sets[i], sl))
+ break;
+ if (l == tl)
+ leaves[tl++] = FcCharSetLeaf(sets[i], sl);
+ }
+ }
+
+ /*
+ * Scan the input until the marker is found
+ */
+
+ while (fgets (line, sizeof (line), stdin))
+ {
+ if (!strncmp (line, "@@@", 3))
+ break;
+ fputs (line, stdout);
+ }
+
+ printf ("/* total size: %d unique leaves: %d */\n\n",
+ total_leaves, tl);
+
+ /*
+ * Find duplicate charsets
+ */
+ duplicate[0] = -1;
+ for (i = 1; sets[i]; i++)
+ {
+ int j;
+
+ duplicate[i] = -1;
+ for (j = 0; j < i; j++)
+ if (sets[j] == sets[i])
+ {
+ duplicate[i] = j;
+ break;
+ }
+ }
+
+ tn = 0;
+ for (i = 0; sets[i]; i++) {
+ if (duplicate[i] >= 0)
+ continue;
+ off[i] = tn;
+ tn += sets[i]->num;
+ }
+
+ printf ("#define LEAF0 (%d * sizeof (FcLangCharSet))\n", nsets);
+ printf ("#define OFF0 (LEAF0 + %d * sizeof (FcCharLeaf))\n", tl);
+ printf ("#define NUM0 (OFF0 + %d * sizeof (intptr_t))\n", tn);
+ printf ("#define SET(n) (n * sizeof (FcLangCharSet) + offsetof (FcLangCharSet, charset))\n");
+ printf ("#define OFF(s,o) (OFF0 + o * sizeof (intptr_t) - SET(s))\n");
+ printf ("#define NUM(s,n) (NUM0 + n * sizeof (FcChar16) - SET(s))\n");
+ printf ("#define LEAF(o,l) (LEAF0 + l * sizeof (FcCharLeaf) - (OFF0 + o * sizeof (intptr_t)))\n");
+ printf ("#define fcLangCharSets (fcLangData.langCharSets)\n");
+ printf ("#define fcLangCharSetIndices (fcLangData.langIndices)\n");
+ printf ("#define fcLangCharSetIndicesInv (fcLangData.langIndicesInv)\n");
+ printf ("\n");
+
+ printf ("static const struct {\n"
+ " FcLangCharSet langCharSets[%d];\n"
+ " FcCharLeaf leaves[%d];\n"
+ " intptr_t leaf_offsets[%d];\n"
+ " FcChar16 numbers[%d];\n"
+ " FcChar%s langIndices[%d];\n"
+ " FcChar%s langIndicesInv[%d];\n"
+ "} fcLangData = {\n",
+ nsets, tl, tn, tn,
+ nsets < 256 ? "8 " : "16", nsets, nsets < 256 ? "8 " : "16", nsets);
+
+ /*
+ * Dump sets
+ */
+
+ printf ("{\n");
+ for (i = 0; sets[i]; i++)
+ {
+ int j = duplicate[i];
+
+ if (j < 0)
+ j = i;
+
+ printf (" { \"%s\", "
+ " { FC_REF_CONSTANT, %d, OFF(%d,%d), NUM(%d,%d) } }, /* %d */\n",
+ langs[i],
+ sets[j]->num, i, off[j], i, off[j], i);
+ }
+ printf ("},\n");
+
+ /*
+ * Dump leaves
+ */
+ printf ("{\n");
+ for (l = 0; l < tl; l++)
+ {
+ printf (" { { /* %d */", l);
+ for (i = 0; i < 256/32; i++)
+ {
+ if (i % 4 == 0)
+ printf ("\n ");
+ printf (" 0x%08x,", leaves[l]->map[i]);
+ }
+ printf ("\n } },\n");
+ }
+ printf ("},\n");
+
+ /*
+ * Dump leaves
+ */
+ printf ("{\n");
+ for (i = 0; sets[i]; i++)
+ {
+ int n;
+
+ if (duplicate[i] >= 0)
+ continue;
+ printf (" /* %s */\n", names[i]);
+ for (n = 0; n < sets[i]->num; n++)
+ {
+ if (n % 4 == 0)
+ printf (" ");
+ for (l = 0; l < tl; l++)
+ if (leaves[l] == FcCharSetLeaf(sets[i], n))
+ break;
+ if (l == tl)
+ fatal (names[i], 0, "can't find leaf");
+ printf (" LEAF(%3d,%3d),", off[i], l);
+ if (n % 4 == 3)
+ printf ("\n");
+ }
+ if (n % 4 != 0)
+ printf ("\n");
+ }
+ printf ("},\n");
+
+
+ printf ("{\n");
+ for (i = 0; sets[i]; i++)
+ {
+ int n;
+
+ if (duplicate[i] >= 0)
+ continue;
+ printf (" /* %s */\n", names[i]);
+ for (n = 0; n < sets[i]->num; n++)
+ {
+ if (n % 8 == 0)
+ printf (" ");
+ printf (" 0x%04x,", FcCharSetNumbers (sets[i])[n]);
+ if (n % 8 == 7)
+ printf ("\n");
+ }
+ if (n % 8 != 0)
+ printf ("\n");
+ }
+ printf ("},\n");
+
+ /* langIndices */
+ printf ("{\n");
+ for (i = 0; sets[i]; i++)
+ {
+ printf (" %d, /* %s */\n", entries[i].id, names[i]);
+ }
+ printf ("},\n");
+
+ /* langIndicesInv */
+ printf ("{\n");
+ {
+ static int entries_inv[MAX_LANG];
+ for (i = 0; sets[i]; i++)
+ entries_inv[entries[i].id] = i;
+ for (i = 0; sets[i]; i++)
+ printf (" %d, /* %s */\n", entries_inv[i], names[entries_inv[i]]);
+ }
+ printf ("}\n");
+
+ printf ("};\n\n");
+
+ printf ("#define NUM_LANG_CHAR_SET %d\n", i);
+ num_lang_set_map = (i + 31) / 32;
+ printf ("#define NUM_LANG_SET_MAP %d\n", num_lang_set_map);
+ /*
+ * Dump indices with country codes
+ */
+ if (ncountry)
+ {
+ int c;
+ int ncountry_ent = 0;
+ printf ("\n");
+ printf ("static const FcChar32 fcLangCountrySets[][NUM_LANG_SET_MAP] = {\n");
+ for (c = 0; c < ncountry; c++)
+ {
+ i = country[c];
+ if (i >= 0)
+ {
+ int lang = strchr (langs[i], '-') - langs[i];
+ int d, k;
+
+ for (k = 0; k < num_lang_set_map; k++)
+ map[k] = 0;
+
+ BitSet (map, i);
+ for (d = c + 1; d < ncountry; d++)
+ {
+ int j = country[d];
+ if (j >= 0 && !strncmp (langs[j], langs[i], lang + 1))
+ {
+ BitSet(map, j);
+ country[d] = -1;
+ }
+ }
+ printf (" {");
+ for (k = 0; k < num_lang_set_map; k++)
+ printf (" 0x%08x,", map[k]);
+ printf (" }, /* %*.*s */\n",
+ lang, lang, langs[i]);
+ ++ncountry_ent;
+ }
+ }
+ printf ("};\n\n");
+ printf ("#define NUM_COUNTRY_SET %d\n", ncountry_ent);
+ }
+
+
+ /*
+ * Find ranges for each letter for faster searching
+ */
+ setRangeChar = 'a';
+ memset(setRangeStart, '\0', sizeof (setRangeStart));
+ memset(setRangeEnd, '\0', sizeof (setRangeEnd));
+ for (i = 0; sets[i]; i++)
+ {
+ char c = names[i][0];
+
+ while (setRangeChar <= c && c <= 'z')
+ setRangeStart[setRangeChar++ - 'a'] = i;
+ }
+ for (setRangeChar = 'a'; setRangeChar < 'z'; setRangeChar++)
+ setRangeEnd[setRangeChar - 'a'] = setRangeStart[setRangeChar+1-'a'] - 1;
+ setRangeEnd[setRangeChar - 'a'] = i - 1;
+
+ /*
+ * Dump sets start/finish for the fastpath
+ */
+ printf ("\n");
+ printf ("static const FcLangCharSetRange fcLangCharSetRanges[] = {\n");
+ printf ("\n");
+ for (setRangeChar = 'a'; setRangeChar <= 'z' ; setRangeChar++)
+ {
+ printf (" { %d, %d }, /* %c */\n",
+ setRangeStart[setRangeChar - 'a'],
+ setRangeEnd[setRangeChar - 'a'], setRangeChar);
+ }
+ printf ("};\n\n");
+
+ while (fgets (line, sizeof (line), stdin))
+ fputs (line, stdout);
+
+ fflush (stdout);
+ exit (ferror (stdout));
+}