[Rework] Use a special structure for stats tokens

src/libmime/message.c

@@ -204,7 +204,7 @@ rspamd_extract_words (struct rspamd_task *task,
 #ifdef WITH_SNOWBALL
 	struct sb_stemmer *stem = NULL;
 #endif
-	rspamd_ftok_t *w;
+	rspamd_stat_token_t *w;
 	gchar *temp_word;
 	const guchar *r;
 	guint i, nlen;
@@ -231,7 +231,7 @@ rspamd_extract_words (struct rspamd_task *task,
 		for (i = 0; i < part->normalized_words->len; i ++) {
 			guint64 h;
 
-			w = &g_array_index (part->normalized_words, rspamd_ftok_t, i);
+			w = &g_array_index (part->normalized_words, rspamd_stat_token_t, i);
 			r = NULL;
 #ifdef WITH_SNOWBALL
 			if (stem) {
@@ -239,7 +239,7 @@ rspamd_extract_words (struct rspamd_task *task,
 			}
 #endif
 
-			if (w->len > 0 && !(w->len == 6 && memcmp (w->begin, "!!EX!!", 6) == 0)) {
+			if (w->len > 0 && (w->flags & RSPAMD_STAT_TOKEN_FLAG_TEXT)) {
 				if (r != NULL) {
 					nlen = strlen (r);
 					nlen = MIN (nlen, w->len);
@@ -268,7 +268,8 @@ rspamd_extract_words (struct rspamd_task *task,
 				 * We use static hash seed if we would want to use that in shingles
 				 * computation in future
 				 */
-				h = rspamd_cryptobox_fast_hash_specific (RSPAMD_CRYPTOBOX_HASHFAST_INDEPENDENT,
+				h = rspamd_cryptobox_fast_hash_specific (
+						RSPAMD_CRYPTOBOX_HASHFAST_INDEPENDENT,
 						w->begin, w->len, words_hash_seed);
 				g_array_append_val (part->normalized_hashes, h);
 			}

src/libstat/stat_api.h

@@ -26,6 +26,18 @@
  * High level statistics API
  */
 
+#define RSPAMD_STAT_TOKEN_FLAG_TEXT (1 << 0)
+#define RSPAMD_STAT_TOKEN_FLAG_META (1 << 1)
+#define RSPAMD_STAT_TOKEN_FLAG_LUA_META (1 << 2)
+#define RSPAMD_STAT_TOKEN_FLAG_EXCEPTION (1 << 3)
+#define RSPAMD_STAT_TOKEN_FLAG_SUBJECT (1 << 4)
+
+typedef struct rspamd_stat_token_s {
+	const gchar *begin;
+	gsize len;
+	guint flags;
+} rspamd_stat_token_t;
+
 /**
  * The results of statistics processing:
  * - error

src/libstat/stat_internal.h

@@ -56,6 +56,7 @@ typedef struct token_node_s {
 	guchar data[RSPAMD_MAX_TOKEN_LEN];
 	guint window_idx;
 	guint datalen;
+	guint flags;
 	gdouble values[];
 } rspamd_token_t;
 

src/libstat/stat_process.c

@@ -38,9 +38,10 @@ rspamd_stat_tokenize_header (struct rspamd_task *task,
 	struct rspamd_mime_header *cur;
 	GPtrArray *hdrs;
 	guint i;
-	rspamd_ftok_t str;
+	rspamd_stat_token_t str;
 
 	hdrs = g_hash_table_lookup (task->raw_headers, name);
+	str.flags = RSPAMD_STAT_TOKEN_FLAG_META;
 
 	if (hdrs != NULL) {
 
@@ -75,12 +76,13 @@ rspamd_stat_tokenize_parts_metadata (struct rspamd_stat_ctx *st_ctx,
 	struct rspamd_mime_text_part *tp;
 	GList *cur;
 	GArray *ar;
-	rspamd_ftok_t elt;
+	rspamd_stat_token_t elt;
 	guint i;
 	gchar tmpbuf[128];
 	lua_State *L = task->cfg->lua_state;
 
 	ar = g_array_sized_new (FALSE, FALSE, sizeof (elt), 16);
+	elt.flags = RSPAMD_STAT_TOKEN_FLAG_META;
 
 	/* Insert images */
 	for (i = 0; i < task->parts->len; i ++) {
@@ -171,6 +173,7 @@ rspamd_stat_tokenize_parts_metadata (struct rspamd_stat_ctx *st_ctx,
 
 	/* Use global metatokens from lua */
 	lua_getglobal (L, "rspamd_gen_metatokens");
+	elt.flags |= RSPAMD_STAT_TOKEN_FLAG_LUA_META;
 
 	if (lua_type (L, -1) == LUA_TFUNCTION) {
 		struct rspamd_task **ptask;
@@ -227,6 +230,7 @@ rspamd_stat_process_tokenize (struct rspamd_stat_ctx *st_ctx,
 		struct rspamd_task *task)
 {
 	struct rspamd_mime_text_part *part;
+	rspamd_stat_token_t *tok;
 	GArray *words;
 	gchar *sub = NULL;
 	guint i, reserved_len = 0;
@@ -272,6 +276,12 @@ rspamd_stat_process_tokenize (struct rspamd_stat_ctx *st_ctx,
 		words = rspamd_tokenize_text (sub, strlen (sub), TRUE, NULL, NULL, FALSE,
 				NULL);
 		if (words != NULL) {
+
+			for (i = 0; i < words->len; i ++) {
+				tok = &g_array_index (words, rspamd_stat_token_t, i);
+				tok->flags |= RSPAMD_STAT_TOKEN_FLAG_SUBJECT;
+			}
+
 			st_ctx->tokenizer->tokenize_func (st_ctx,
 					task->task_pool,
 					words,

src/libstat/tokenizers/osb.c

@@ -264,12 +264,12 @@ rspamd_tokenizer_osb (struct rspamd_stat_ctx *ctx,
 		GPtrArray *result)
 {
 	rspamd_token_t *new_tok = NULL;
-	rspamd_ftok_t *token;
+	rspamd_stat_token_t *token;
 	struct rspamd_osb_tokenizer_config *osb_cf;
 	guint64 *hashpipe, cur, seed;
 	guint32 h1, h2;
 	gsize token_size;
-	guint processed = 0, i, w, window_size;
+	guint processed = 0, i, w, window_size, token_flags = 0;
 
 	if (words == NULL) {
 		return FALSE;
@@ -292,10 +292,15 @@ rspamd_tokenizer_osb (struct rspamd_stat_ctx *ctx,
 	g_assert (token_size > 0);
 
 	for (w = 0; w < words->len; w ++) {
-		token = &g_array_index (words, rspamd_ftok_t, w);
+		token = &g_array_index (words, rspamd_stat_token_t, w);
+		token_flags = token->flags;
 
 		if (osb_cf->ht == RSPAMD_OSB_HASH_COMPAT) {
-			cur = rspamd_fstrhash_lc (token, is_utf);
+			rspamd_ftok_t ftok;
+
+			ftok.begin = token->begin;
+			ftok.len = token->len;
+			cur = rspamd_fstrhash_lc (&ftok, is_utf);
 		}
 		else {
 			/* We know that the words are normalized */
@@ -316,6 +321,7 @@ rspamd_tokenizer_osb (struct rspamd_stat_ctx *ctx,
 #define ADD_TOKEN do {\
     new_tok = rspamd_mempool_alloc0 (pool, token_size); \
     new_tok->datalen = sizeof (gint64); \
+    new_tok->flags = token_flags; \
     if (osb_cf->ht == RSPAMD_OSB_HASH_COMPAT) { \
         h1 = ((guint32)hashpipe[0]) * primes[0] + \
             ((guint32)hashpipe[i]) * primes[i << 1]; \

src/libstat/tokenizers/tokenizers.c

@@ -22,8 +22,8 @@
 #include "stat_internal.h"
 #include "../../../contrib/mumhash/mum.h"
 
-typedef gboolean (*token_get_function) (rspamd_ftok_t * buf, gchar const **pos,
-		rspamd_ftok_t * token,
+typedef gboolean (*token_get_function) (rspamd_stat_token_t * buf, gchar const **pos,
+		rspamd_stat_token_t * token,
 		GList **exceptions, gboolean is_utf, gsize *rl, gboolean check_signature);
 
 const gchar t_delimiters[255] = {
@@ -69,8 +69,8 @@ token_node_compare_func (gconstpointer a, gconstpointer b)
 
 /* Get next word from specified f_str_t buf */
 static gboolean
-rspamd_tokenizer_get_word_compat (rspamd_ftok_t * buf,
-		gchar const **cur, rspamd_ftok_t * token,
+rspamd_tokenizer_get_word_compat (rspamd_stat_token_t * buf,
+		gchar const **cur, rspamd_stat_token_t * token,
 		GList **exceptions, gboolean is_utf, gsize *rl, gboolean unused)
 {
 	gsize remain, pos;
@@ -92,6 +92,7 @@ rspamd_tokenizer_get_word_compat (rspamd_ftok_t * buf,
 			if (ex->pos == 0) {
 				token->begin = buf->begin + ex->len;
 				token->len = ex->len;
+				token->flags = RSPAMD_STAT_TOKEN_FLAG_EXCEPTION;
 			}
 			else {
 				token->begin = buf->begin;
@@ -155,14 +156,16 @@ rspamd_tokenizer_get_word_compat (rspamd_ftok_t * buf,
 		}
 	}
 
+	token->flags = RSPAMD_STAT_TOKEN_FLAG_TEXT;
+
 	*cur = p;
 
 	return TRUE;
 }
 
 static gboolean
-rspamd_tokenizer_get_word (rspamd_ftok_t * buf,
-		gchar const **cur, rspamd_ftok_t * token,
+rspamd_tokenizer_get_word (rspamd_stat_token_t * buf,
+		gchar const **cur, rspamd_stat_token_t * token,
 		GList **exceptions, gboolean is_utf, gsize *rl,
 		gboolean check_signature)
 {
@@ -219,6 +222,7 @@ rspamd_tokenizer_get_word (rspamd_ftok_t * buf,
 				if (ex->type == RSPAMD_EXCEPTION_URL) {
 					token->begin = "!!EX!!";
 					token->len = sizeof ("!!EX!!") - 1;
+					token->flags = RSPAMD_STAT_TOKEN_FLAG_EXCEPTION;
 					processed = token->len;
 				}
 				state = skip_exception;
@@ -240,9 +244,11 @@ rspamd_tokenizer_get_word (rspamd_ftok_t * buf,
 			break;
 		case feed_token:
 			if (ex != NULL && p - buf->begin == (gint)ex->pos) {
+				token->flags = RSPAMD_STAT_TOKEN_FLAG_TEXT;
 				goto set_token;
 			}
 			else if (!g_unichar_isgraph (uc) || g_unichar_ispunct (uc)) {
+				token->flags = RSPAMD_STAT_TOKEN_FLAG_TEXT;
 				goto set_token;
 			}
 			processed ++;
@@ -288,7 +294,7 @@ rspamd_tokenize_text (gchar *text, gsize len, gboolean is_utf,
 		struct rspamd_config *cfg, GList *exceptions, gboolean compat,
 		guint64 *hash)
 {
-	rspamd_ftok_t token, buf;
+	rspamd_stat_token_t token, buf;
 	const gchar *pos = NULL;
 	gsize l;
 	GArray *res;
@@ -322,7 +328,8 @@ rspamd_tokenize_text (gchar *text, gsize len, gboolean is_utf,
 		initial_size = word_decay * 2;
 	}
 
-	res = g_array_sized_new (FALSE, FALSE, sizeof (rspamd_ftok_t), initial_size);
+	res = g_array_sized_new (FALSE, FALSE, sizeof (rspamd_stat_token_t),
+			initial_size);
 
 	while (func (&buf, &pos, &token, &cur, is_utf, &l, FALSE)) {
 		if (l == 0 || (min_len > 0 && l < min_len) ||

src/libstat/tokenizers/tokenizers.h

@@ -29,7 +29,7 @@ struct rspamd_stat_tokenizer {
 gint token_node_compare_func (gconstpointer a, gconstpointer b);
 
 
-/* Tokenize text into array of words (rspamd_ftok_t type) */
+/* Tokenize text into array of words (rspamd_stat_token_t type) */
 GArray * rspamd_tokenize_text (gchar *text, gsize len, gboolean is_utf,
 		struct rspamd_config *cfg, GList *exceptions, gboolean compat,
 		guint64 *hash);

src/lua/lua_util.c

@@ -895,7 +895,7 @@ lua_util_tokenize_text (lua_State *L)
 	struct rspamd_lua_text *t;
 	struct rspamd_process_exception *ex;
 	GArray *res;
-	rspamd_ftok_t *w;
+	rspamd_stat_token_t *w;
 	gboolean compat = FALSE;
 
 	if (lua_type (L, 1) == LUA_TSTRING) {
@@ -959,7 +959,7 @@ lua_util_tokenize_text (lua_State *L)
 		lua_createtable (L, res->len, 0);
 
 		for (i = 0; i < res->len; i ++) {
-			w = &g_array_index (res, rspamd_ftok_t, i);
+			w = &g_array_index (res, rspamd_stat_token_t, i);
 			lua_pushlstring (L, w->begin, w->len);
 			lua_rawseti (L, -2, i + 1);
 		}

src/plugins/chartable.c

@@ -25,6 +25,7 @@
 #include "config.h"
 #include "libmime/message.h"
 #include "rspamd.h"
+#include "libstat/stat_api.h"
 
 #define DEFAULT_SYMBOL "R_MIXED_CHARSET"
 #define DEFAULT_URL_SYMBOL "R_MIXED_CHARSET_URL"
@@ -163,7 +164,8 @@ chartable_module_reconfig (struct rspamd_config *cfg)
 }
 
 static gdouble
-rspamd_chartable_process_word_utf (struct rspamd_task *task, rspamd_ftok_t *w,
+rspamd_chartable_process_word_utf (struct rspamd_task *task,
+		rspamd_stat_token_t *w,
 		gboolean is_url)
 {
 	const gchar *p, *end, *c;
@@ -258,7 +260,8 @@ rspamd_chartable_process_word_utf (struct rspamd_task *task, rspamd_ftok_t *w,
 }
 
 static gdouble
-rspamd_chartable_process_word_ascii (struct rspamd_task *task, rspamd_ftok_t *w,
+rspamd_chartable_process_word_ascii (struct rspamd_task *task,
+		rspamd_stat_token_t *w,
 		gboolean is_url)
 {
 	const guchar *p, *end, *c;
@@ -343,7 +346,7 @@ static void
 rspamd_chartable_process_part (struct rspamd_task *task,
 		struct rspamd_mime_text_part *part)
 {
-	rspamd_ftok_t *w;
+	rspamd_stat_token_t *w;
 	guint i;
 	gdouble cur_score = 0.0;
 
@@ -353,9 +356,9 @@ rspamd_chartable_process_part (struct rspamd_task *task,
 	}
 
 	for (i = 0; i < part->normalized_words->len; i++) {
-		w = &g_array_index (part->normalized_words, rspamd_ftok_t, i);
+		w = &g_array_index (part->normalized_words, rspamd_stat_token_t, i);
 
-		if (w->len > 0) {
+		if (w->len > 0 && (w->flags & RSPAMD_STAT_TOKEN_FLAG_TEXT)) {
 
 			if (IS_PART_UTF (part)) {
 				cur_score += rspamd_chartable_process_word_utf (task, w, FALSE);
@@ -397,7 +400,7 @@ chartable_url_symbol_callback (struct rspamd_task *task, void *unused)
 	struct rspamd_url *u;
 	GHashTableIter it;
 	gpointer k, v;
-	rspamd_ftok_t w;
+	rspamd_stat_token_t w;
 	gdouble cur_score = 0.0;
 
 	g_hash_table_iter_init (&it, task->urls);

src/plugins/fuzzy_check.c

@@ -43,6 +43,7 @@
 #include "lua/lua_common.h"
 #include "unix-std.h"
 #include "libutil/http_private.h"
+#include "libstat/stat_api.h"
 #include <math.h>
 
 #define DEFAULT_SYMBOL "R_FUZZY_HASH"
@@ -1266,7 +1267,7 @@ fuzzy_cmd_from_text_part (struct fuzzy_rule *rule,
 	struct rspamd_shingle *sh;
 	guint i;
 	rspamd_cryptobox_hash_state_t st;
-	rspamd_ftok_t *word;
+	rspamd_stat_token_t *word;
 	GArray *words;
 	struct fuzzy_cmd_io *io;
 
@@ -1289,7 +1290,7 @@ fuzzy_cmd_from_text_part (struct fuzzy_rule *rule,
 		words = fuzzy_preprocess_words (part, pool);
 
 		for (i = 0; i < words->len; i ++) {
-			word = &g_array_index (words, rspamd_ftok_t, i);
+			word = &g_array_index (words, rspamd_stat_token_t, i);
 			rspamd_cryptobox_hash_update (&st, word->begin, word->len);
 		}
 		rspamd_cryptobox_hash_final (&st, shcmd->basic.digest);