charsets.c

一个很有名的浏览器

	if (first_time) {
		memset(&nb_entity_cache, 0, ENTITY_CACHE_MAXLEN * sizeof(unsigned int));
		first_time = 0;
	}

	/* Check if cached. A test on many websites (freshmeat.net + whole ELinks website
	 * + google + slashdot + websites that result from a search for test on google,
	 * + various ones) show a quite impressive improvment:
	 * Top ten is:
	 * 0: hits=2459 l=4 st='nbsp'
	 * 1: hits=2152 l=6 st='eacute'
	 * 2: hits=235 l=6 st='egrave'
	 * 3: hits=136 l=6 st='agrave'
	 * 4: hits=100 l=3 st='amp'
	 * 5: hits=40 l=5 st='laquo'
	 * 6: hits=8 l=4 st='copy'
	 * 7: hits=5 l=2 st='gt'
	 * 8: hits=2 l=2 st='lt'
	 * 9: hits=1 l=6 st='middot'
	 *
	 * Most of the time cache hit ratio is near 95%.
	 *
	 * A long test shows: 15186 hits vs. 24 misses and mean iteration
	 * count is kept < 2 (worst case 1.58). Not so bad ;)
	 *
	 * --Zas */

	/* entities with length >= ENTITY_CACHE_MAXLEN or == 1 will go in [0] table */
	slen = (strlen > 1 && strlen < ENTITY_CACHE_MAXLEN) ? strlen : 0;

	if (strlen < ENTITY_CACHE_MAXLEN && nb_entity_cache[slen] > 0) {
		int i;

		for (i = 0; i < nb_entity_cache[slen]; i++) {
			if (entity_cache[slen][i].encoding == encoding
			    && !memcmp(str, entity_cache[slen][i].str, strlen)) {
#ifdef DEBUG_ENTITY_CACHE
				static double total_iter = 0;
				static unsigned long hit_count = 0;

				total_iter += i + 1;
				hit_count++;
				fprintf(stderr, "hit after %d iter. (mean = %0.2f)\n", i + 1, total_iter / (double) hit_count);
#endif
				if (entity_cache[slen][i].hits < (unsigned int) ~0)
					entity_cache[slen][i].hits++;
				return entity_cache[slen][i].result;
			}
		}
#ifdef DEBUG_ENTITY_CACHE
		fprintf(stderr, "miss\n");
#endif
	}

	if (*str == '#') { /* Numeric entity. */
		int l = (int) strlen;
		unsigned char *st = (unsigned char *) str;
		unicode_val n = 0;

		if (l == 1) goto end; /* &#; ? */
		st++, l--;
		if ((*st | 32) == 'x') { /* Hexadecimal */

			if (l == 1 || l > 9) goto end; /* xFFFFFFFF max. */
			st++, l--;
			do {
				unsigned char c = (*(st++) | 32);

				if (isdigit(c))
					n = (n << 4) | (c - '0');
				else if (isxdigit(c))
					n = (n << 4) | (c - 'a' + 10);
				else
					goto end; /* Bad char. */
			} while (--l);
		} else { /* Decimal */
			if (l > 10) goto end; /* 4294967295 max. */
			do {
				unsigned char c = *(st++);

				if (isdigit(c))
					n = n * 10 + c - '0';
				else
					goto end; /* Bad char. */
				/* Limit to 0xFFFFFFFF. */
				if (n == (unicode_val) 0xFFFFFFFF)
					goto end;
			} while (--l);
		}

		result = u2cp(n, encoding);

#ifdef DEBUG_ENTITY_CACHE
		fprintf(stderr, "%lu %016x %s\n", (unsigned long) n , n, result);
#endif
	} else { /* Text entity. */
		struct string key = INIT_STRING((unsigned char *) str, strlen);
		struct entity *element = bsearch((void *) &key, entities,
						 N_ENTITIES,
						 sizeof(*element),
						 compare_entities);

		if (element) result = u2cp(element->c, encoding);
	}

end:
	/* Take care of potential buffer overflow. */
	if (strlen < sizeof(entity_cache[slen][0].str)) {
		struct entity_cache *ece = &entity_cache[slen][nb_entity_cache[slen]];

		/* Copy new entry to cache. */
		ece->hits = 1;
		ece->strlen = strlen;
		ece->encoding = encoding;
		ece->result = result;
		memcpy(ece->str, str, strlen);
		ece->str[strlen] = '\0';

		/* Increment number of cache entries if possible. */
		if (nb_entity_cache[slen] < ENTITY_CACHE_SIZE) nb_entity_cache[slen]++;

#ifdef DEBUG_ENTITY_CACHE
		fprintf(stderr, "Added in [%d]: l=%d st='%s'\n", slen,
				entity_cache[slen][0].strlen, entity_cache[slen][0].str);

#endif

		/* Sort entries by hit order. */
		if (nb_entity_cache[slen] > 1)
			qsort(&entity_cache[slen][0], nb_entity_cache[slen],
			      sizeof(entity_cache[slen][0]), (void *) hits_cmp);

#ifdef DEBUG_ENTITY_CACHE
	{
		unsigned int i;

		fprintf(stderr, "- Cache entries [%d] -\n", slen);
		for (i = 0; i < nb_entity_cache[slen] ; i++)
			fprintf(stderr, "%d: hits=%d l=%d st='%s'\n", i,
				entity_cache[slen][i].hits, entity_cache[slen][i].strlen,
				entity_cache[slen][i].str);
		fprintf(stderr, "-----------------\n");
	}
#endif
	}
	return result;
}

unsigned char *
convert_string(struct conv_table *convert_table,
	       unsigned char *chars, int charslen,
	       enum convert_string_mode mode, int *length,
	       void (*callback)(void *data, unsigned char *buf, int buflen),
	       void *callback_data)
{
	unsigned char *buffer;
	int bufferpos = 0;
	int charspos = 0;

	if (!convert_table && !memchr(chars, '&', charslen)) {
		if (callback) {
			if (charslen) callback(callback_data, chars, charslen);
			return NULL;
		} else {
			return memacpy(chars, charslen);
		}
	}

	/* Buffer allocation */

	buffer = mem_alloc(ALLOC_GR + 1 /* trailing \0 */);
	if (!buffer) return NULL;

	/* Iterate ;-) */

	while (charspos < charslen) {
		unsigned char *translit;

#define PUTC do { \
		buffer[bufferpos++] = chars[charspos++]; \
		translit = ""; \
		goto flush; \
	} while (0)

		if (chars[charspos] != '&') {
			struct conv_table *t;
			int i;

			if (chars[charspos] < 128 || !convert_table) PUTC;

			t = convert_table;
			i = charspos;

			while (t[chars[i]].t) {
				t = t[chars[i++]].u.tbl;
				if (i >= charslen) PUTC;
			}

			translit = t[chars[i]].u.str;
			charspos = i + 1;

		} else if (mode == CSM_FORM || mode == CSM_NONE) {
			PUTC;

		} else {
			int start = charspos + 1;
			int i = start;

			while (i < charslen
			       && (isasciialpha(chars[i])
				   || isdigit(chars[i])
				   || (chars[i] == '#')))
				i++;

			/* This prevents bug 213: we were expanding "entities"
			 * in URL query strings. */
			/* XXX: But this disables &nbsp&nbsp usage, which
			 * appears to be relatively common! --pasky */
			if ((mode == CSM_DEFAULT || (chars[i] != '&' && chars[i] != '='))
			    && i > start
			    && !isasciialpha(chars[i]) && !isdigit(chars[i])) {
				translit = get_entity_string(&chars[start], i - start,
						      global_doc_opts->cp);
				if (chars[i] != ';') {
					/* Eat &nbsp &nbsp<foo> happily, but
					 * pull back from the character after
					 * entity string if it is not the valid
					 * terminator. */
					i--;
				}

				if (!translit) PUTC;
				charspos = i + (i < charslen);
			} else PUTC;
		}

		if (!translit[0]) continue;

		if (!translit[1]) {
			buffer[bufferpos++] = translit[0];
			translit = "";
			goto flush;
		}

		while (*translit) {
			unsigned char *new;

			buffer[bufferpos++] = *(translit++);
flush:
			if (bufferpos & (ALLOC_GR - 1)) continue;

			if (callback) {
				buffer[bufferpos] = 0;
				callback(callback_data, buffer, bufferpos);
				bufferpos = 0;
			} else {
				new = mem_realloc(buffer, bufferpos + ALLOC_GR);
				if (!new) {
					mem_free(buffer);
					return NULL;
				}
				buffer = new;
			}
		}
#undef PUTC
	}

	/* Say bye */

	buffer[bufferpos] = 0;
	if (length) *length = bufferpos;

	if (callback) {
		if (bufferpos) callback(callback_data, buffer, bufferpos);
		mem_free(buffer);
		return NULL;
	} else {
		return buffer;
	}
}


#ifndef USE_FASTFIND
int
get_cp_index(unsigned char *name)
{
	int i, a;
	int syscp = 0;

	if (!strcasecmp(name, "System")) {
#if HAVE_LANGINFO_CODESET
		name = nl_langinfo(CODESET);
		syscp = SYSTEM_CHARSET_FLAG;
#else
		name = "us-ascii";
#endif
	}

	for (i = 0; codepages[i].name; i++) {
		for (a = 0; codepages[i].aliases[a]; a++) {
			/* In the past, we looked for the longest substring
			 * in all the names; it is way too expensive, though:
			 *
			 *   %   cumulative   self              self     total
			 *  time   seconds   seconds    calls  us/call  us/call  name
			 *  3.00      0.66     0.03     1325    22.64    22.64  get_cp_index
			 *
			 * Anything called from redraw_screen() is in fact
			 * relatively expensive, even if it's called just
			 * once. So we will do a simple strcasecmp() here.
			 */

			if (!strcasecmp(name, codepages[i].aliases[a]))
				return i | syscp;
		}
	}

	if (syscp) {
		return get_cp_index("us-ascii") | syscp;
	} else {
		return -1;
	}
}

#else

static unsigned int i_name = 0;
static unsigned int i_alias = 0;

/* Reset internal list pointer */
void
charsets_list_reset(void)
{
	i_name = 0;
	i_alias = 0;
}

/* Returns a pointer to a struct that contains current key and data pointers
 * and increment internal pointer.  It returns NULL when key is NULL. */
struct fastfind_key_value *
charsets_list_next(void)
{
	static struct fastfind_key_value kv;

	if (!codepages[i_name].name) return NULL;

	kv.key = codepages[i_name].aliases[i_alias];
	kv.data = &codepages[i_name];

	if (codepages[i_name].aliases[i_alias + 1])
		i_alias++;
	else {
		i_name++;
		i_alias = 0;
	}

	return &kv;
}

static struct fastfind_index ff_charsets_index
	= INIT_FASTFIND_INDEX("charsets_lookup", charsets_list_reset, charsets_list_next);

/* It searchs for a charset named @name or one of its aliases and
 * returns index for it or -1 if not found. */
int
get_cp_index(unsigned char *name)
{
	struct codepage_desc *codepage;
	int syscp = 0;

	if (!strcasecmp(name, "System")) {
#if HAVE_LANGINFO_CODESET
		name = nl_langinfo(CODESET);
		syscp = SYSTEM_CHARSET_FLAG;
#else
		name = "us-ascii";
#endif
	}

	codepage = fastfind_search(&ff_charsets_index, name, strlen(name));
	if (codepage) {
		assert(codepages <= codepage && codepage < codepages + N_CODEPAGES);
		return (codepage - codepages) | syscp;

	} else if (syscp) {
		return get_cp_index("us-ascii") | syscp;

	} else {
		return -1;
	}
}

#endif /* USE_FASTFIND */

void
init_charsets_lookup(void)
{
#ifdef USE_FASTFIND
	fastfind_index(&ff_charsets_index, FF_COMPRESS);
#endif
}

void
free_charsets_lookup(void)
{
#ifdef USE_FASTFIND
	fastfind_done(&ff_charsets_index);
#endif
}

unsigned char *
get_cp_name(int cp_index)
{
	if (cp_index < 0) return "none";
	if (cp_index & SYSTEM_CHARSET_FLAG) return "System";

	return codepages[cp_index].name;
}

unsigned char *
get_cp_mime_name(int cp_index)
{
	if (cp_index < 0) return "none";
	if (cp_index & SYSTEM_CHARSET_FLAG) return "System";
	if (!codepages[cp_index].aliases) return NULL;

	return codepages[cp_index].aliases[0];
}

int
is_cp_special(int cp_index)
{
	cp_index &= ~SYSTEM_CHARSET_FLAG;
	return codepages[cp_index].table == table_utf_8;
}