* Add new algorithm based on diff algorithm to compare relatively short text parts

14 files changed, 498 insertions, 11 deletions

diff --git a/lib/CMakeLists.txt b/lib/CMakeLists.txt
index 2bb4e028e..a8d0527f1 100644
--- a/lib/CMakeLists.txt
+++ b/lib/CMakeLists.txt
@@ -28,6 +28,7 @@ SET(RSPAMDLIBSRC ../src/binlog.c
 				../src/buffer.c
 				../src/cfg_utils.c
 				../src/cfg_xml.c
+				../src/diff.c
 				../src/dns.c
 				../src/events.c
 				../src/expressions.c
diff --git a/src/cfg_file.h b/src/cfg_file.h
index 64a46d149..6a4abb9b8 100644
--- a/src/cfg_file.h
+++ b/src/cfg_file.h
@@ -259,6 +259,8 @@ struct config_file {
 	gboolean check_text_attachements;				/**< check text attachements as text					*/
 	gboolean convert_config;						/**< convert config to XML format						*/
 
+	gsize max_diff;									/**< maximum diff size for text parts					*/
+
 	enum rspamd_log_type log_type;					/**< log type											*/
 	gint log_facility;								/**< log facility in case of syslog						*/
 	gint log_level;									/**< log level trigger									*/
diff --git a/src/cfg_utils.c b/src/cfg_utils.c
index 56202d4f1..a34084cce 100644
--- a/src/cfg_utils.c
+++ b/src/cfg_utils.c
@@ -175,6 +175,9 @@ init_defaults (struct config_file *cfg)
 	cfg->statfile_sync_interval = 60000;
 	cfg->statfile_sync_timeout = 20000;
 
+	/* 20 Kb */
+	cfg->max_diff = 20480;
+
 	cfg->max_statfile_size = DEFAULT_STATFILE_SIZE;
 	cfg->modules_opts = g_hash_table_new (g_str_hash, g_str_equal);
 	cfg->variables = g_hash_table_new (g_str_hash, g_str_equal);
diff --git a/src/cfg_xml.c b/src/cfg_xml.c
index bb23dd641..9920f355c 100644
--- a/src/cfg_xml.c
+++ b/src/cfg_xml.c
@@ -293,6 +293,12 @@ static struct xml_parser_rule grammar[] = {
 				G_STRUCT_OFFSET (struct config_file, statfile_sync_timeout),
 				NULL
 			},
+			{
+				"max_diff",
+				xml_handle_size,
+				G_STRUCT_OFFSET (struct config_file, max_diff),
+				NULL
+			},
 			NULL_ATTR
 		},
 		NULL_DEF_ATTR
diff --git a/src/diff.c b/src/diff.c
new file mode 100644
index 000000000..a2e84349b
--- /dev/null
+++ b/src/diff.c
@@ -0,0 +1,379 @@
+/* diff - compute a shortest edit script (SES) given two sequences
+ * Copyright (c) 2004 Michael B. Allen <mba2000 ioplex.com>
+ *
+ * The MIT License
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included
+ * in all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
+ * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ */
+
+/* This algorithm is basically Myers' solution to SES/LCS with
+ * the Hirschberg linear space refinement as described in the
+ * following publication:
+ *
+ *   E. Myers, ``An O(ND) Difference Algorithm and Its Variations,''
+ *   Algorithmica 1, 2 (1986), 251-266.
+ *   http://www.cs.arizona.edu/people/gene/PAPERS/diff.ps
+ *
+ * This is the same algorithm used by GNU diff(1).
+ */
+
+/* Copyright (c) 2010, Vsevolod Stakhov
+ * All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions are met:
+ *       * Redistributions of source code must retain the above copyright
+ *         notice, this list of conditions and the following disclaimer.
+ *       * Redistributions in binary form must reproduce the above copyright
+ *         notice, this list of conditions and the following disclaimer in the
+ *         documentation and/or other materials provided with the distribution.
+ *
+ * THIS SOFTWARE IS PROVIDED ''AS IS'' AND ANY
+ * EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
+ * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
+ * DISCLAIMED. IN NO EVENT SHALL Rambler BE LIABLE FOR ANY
+ * DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
+ * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
+ * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
+ * ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+ * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ */
+
+#include "config.h"
+#include "diff.h"
+
+
+#define FV(k) _v(ctx, (k), 0)
+#define RV(k) _v(ctx, (k), 1)
+
+#define MAX_DIFF 1024
+
+struct _ctx
+{
+	GArray *buf;
+	GArray *ses;
+	gint si;
+	gint dmax;
+};
+
+struct middle_snake
+{
+	gint x, y, u, v;
+};
+
+static
+void maybe_resize_array(GArray *arr, guint k)
+{
+	if (k > arr->len) {
+		g_array_set_size (arr, k);
+	}
+
+}
+
+static void
+_setv(struct _ctx *ctx, gint k, gint r, gint val)
+{
+	gint j;
+	gint *i;
+	/* Pack -N to N ginto 0 to N * 2
+	 */
+	j = k <= 0 ? -k * 4 + r : k * 4 + (r - 2);
+
+	maybe_resize_array (ctx->buf, j);
+	i = (gint *) &g_array_index (ctx->buf, gint, j);
+	*i = val;
+}
+
+static gint
+_v(struct _ctx *ctx, gint k, gint r)
+{
+	gint j;
+
+	j = k <= 0 ? -k * 4 + r : k * 4 + (r - 2);
+
+	return *((gint *) &g_array_index (ctx->buf, gint, j));
+}
+
+static gint
+_find_middle_snake(const void *a, gint aoff, gint n, const void *b,
+		gint boff, gint m, struct _ctx *ctx, struct middle_snake *ms)
+{
+	gint delta, odd, mid, d;
+
+	delta = n - m;
+	odd = delta & 1;
+	mid = (n + m) / 2;
+	mid += odd;
+
+	_setv (ctx, 1, 0, 0);
+	_setv (ctx, delta - 1, 1, n);
+
+	for (d = 0; d <= mid; d++) {
+		gint k, x, y;
+
+		if ((2 * d - 1) >= ctx->dmax) {
+			return ctx->dmax;
+		}
+
+		for (k = d; k >= -d; k -= 2) {
+			if (k == -d || (k != d && FV(k - 1) < FV(k + 1))) {
+				x = FV(k + 1);
+			}
+			else {
+				x = FV(k - 1) + 1;
+			}
+			y = x - k;
+
+			ms->x = x;
+			ms->y = y;
+			const guchar *a0 = (const guchar *) a + aoff;
+			const guchar *b0 = (const guchar *) b + boff;
+			while (x < n && y < m && a0[x] == b0[y]) {
+				x++;
+				y++;
+			}
+			_setv (ctx, k, 0, x);
+
+			if (odd && k >= (delta - (d - 1)) && k <= (delta + (d - 1))) {
+				if (x >= RV(k)) {
+					ms->u = x;
+					ms->v = y;
+					return 2 * d - 1;
+				}
+			}
+		}
+		for (k = d; k >= -d; k -= 2) {
+			gint kr = (n - m) + k;
+
+			if (k == d || (k != -d && RV(kr - 1) < RV(kr + 1))) {
+				x = RV(kr - 1);
+			}
+			else {
+				x = RV(kr + 1) - 1;
+			}
+			y = x - kr;
+
+			ms->u = x;
+			ms->v = y;
+			const guchar *a0 = (const guchar *) a + aoff;
+			const guchar *b0 = (const guchar *) b + boff;
+			while (x > 0 && y > 0 && a0[x - 1] == b0[y - 1]) {
+				x--;
+				y--;
+			}
+			_setv (ctx, kr, 1, x);
+
+			if (!odd && kr >= -d && kr <= d) {
+				if (x <= FV(kr)) {
+					ms->x = x;
+					ms->y = y;
+					return 2 * d;
+				}
+			}
+		}
+	}
+
+	errno = EFAULT;
+
+	return -1;
+}
+
+static void
+_edit(struct _ctx *ctx, gint op, gint off, gint len)
+{
+	struct diff_edit *e = NULL, newe;
+
+	if (len == 0 || ctx->ses == NULL) {
+		return;
+	}
+	/*
+	 * Add an edit to the SES (or
+	 * coalesce if the op is the same)
+	 */
+	if (ctx->ses->len != 0) {
+		e = &g_array_index (ctx->ses, struct diff_edit, ctx->ses->len - 1);
+	}
+	if (e == NULL || e->op != op) {
+		newe.op = op;
+		newe.off = off;
+		newe.len = len;
+		g_array_append_val (ctx->ses, newe);
+	}
+	else {
+		e->len += len;
+	}
+}
+
+static gint
+_ses(const void *a, gint aoff, gint n, const void *b, gint boff,
+		gint m, struct _ctx *ctx)
+{
+	struct middle_snake ms;
+	gint d;
+
+	if (n == 0) {
+		_edit (ctx, DIFF_INSERT, boff, m);
+		d = m;
+	}
+	else if (m == 0) {
+		_edit (ctx, DIFF_DELETE, aoff, n);
+		d = n;
+	}
+	else {
+		/* Find the middle "snake" around which we
+		 * recursively solve the sub-problems.
+		 */
+		d = _find_middle_snake (a, aoff, n, b, boff, m, ctx, &ms);
+		if (d == -1) {
+			return -1;
+		}
+		else if (d >= ctx->dmax) {
+			return ctx->dmax;
+		}
+		else if (ctx->ses == NULL) {
+			return d;
+		}
+		else if (d > 1) {
+			if (_ses (a, aoff, ms.x, b, boff, ms.y, ctx) == -1) {
+				return -1;
+			}
+
+			_edit (ctx, DIFF_MATCH, aoff + ms.x, ms.u - ms.x);
+
+			aoff += ms.u;
+			boff += ms.v;
+			n -= ms.u;
+			m -= ms.v;
+			if (_ses (a, aoff, n, b, boff, m, ctx) == -1) {
+				return -1;
+			}
+		}
+		else {
+			gint x = ms.x;
+			gint u = ms.u;
+
+			/* There are only 4 base cases when the
+			 * edit distance is 1.
+			 *
+			 * n > m   m > n
+			 *
+			 *   -       |
+			 *    \       \    x != u
+			 *     \       \
+			 *
+			 *   \       \
+			 *    \       \    x == u
+			 *     -       |
+			 */
+
+			if (m > n) {
+				if (x == u) {
+					_edit (ctx, DIFF_MATCH, aoff, n);
+					_edit (ctx, DIFF_INSERT, boff + (m - 1), 1);
+				}
+				else {
+					_edit (ctx, DIFF_INSERT, boff, 1);
+					_edit (ctx, DIFF_MATCH, aoff, n);
+				}
+			}
+			else {
+				if (x == u) {
+					_edit (ctx, DIFF_MATCH, aoff, m);
+					_edit (ctx, DIFF_DELETE, aoff + (n - 1), 1);
+				}
+				else {
+					_edit (ctx, DIFF_DELETE, aoff, 1);
+					_edit (ctx, DIFF_MATCH, aoff + 1, m);
+				}
+			}
+		}
+	}
+
+	return d;
+}
+
+gint
+rspamd_diff(const void *a, gint aoff, gint n, const void *b, gint boff, gint m,
+		gint dmax, GArray *ses, gint *sn)
+{
+	struct _ctx ctx;
+	gint d, x, y;
+	struct diff_edit *e = NULL;
+	GArray *tmp;
+
+	tmp = g_array_sized_new (FALSE, TRUE, sizeof(gint), dmax);
+	ctx.buf = tmp;
+	ctx.ses = ses;
+	ctx.si = 0;
+	ctx.dmax = dmax;
+
+	/* The _ses function assumes the SES will begin or end with a delete
+	 * or insert. The following will insure this is true by eating any
+	 * beginning matches. This is also a quick to process sequences
+	 * that match entirely.
+	 */
+	x = y = 0;
+	const guchar *a0 = (const guchar *) a + aoff;
+	const guchar *b0 = (const guchar *) b + boff;
+	while (x < n && y < m && a0[x] == b0[y]) {
+		x++;
+		y++;
+	}
+	_edit (&ctx, DIFF_MATCH, aoff, x);
+
+	if ((d = _ses (a, aoff + x, n - x, b, boff + y, m - y, &ctx)) == -1) {
+		g_array_free (tmp, TRUE);
+		return -1;
+	}
+	if (ses && sn) {
+		*sn = e->op ? ctx.si + 1 : 0;
+	}
+
+	g_array_free (tmp, TRUE);
+	return d;
+}
+
+guint32
+compare_diff_distance (f_str_t *s1, f_str_t *s2)
+{
+	GArray *ses;
+	struct diff_edit *e;
+	gint i;
+	guint32 distance = 0;
+
+	ses = g_array_sized_new (FALSE, TRUE, sizeof (struct diff_edit), MAX_DIFF);
+
+	if (rspamd_diff (s1->begin, 0, s1->len,
+				s2->begin, 0, s2->len, MAX_DIFF, ses, NULL) == -1) {
+		/* Diff failed, strings are different */
+		g_array_free (ses, TRUE);
+		return 0;
+	}
+
+	for (i = 0; i < ses->len; i ++) {
+		e = &g_array_index(ses, struct diff_edit, i);
+		if (e->op != DIFF_MATCH) {
+			distance += e->len;
+		}
+	}
+
+	g_array_free (ses, TRUE);
+	return 100 - (2 * distance * 100) / (s1->len + s2->len);
+}
diff --git a/src/diff.h b/src/diff.h
new file mode 100644
index 000000000..b1c5426d2
--- /dev/null
+++ b/src/diff.h
@@ -0,0 +1,49 @@
+/* Copyright (c) 2010, Vsevolod Stakhov
+ * All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions are met:
+ *       * Redistributions of source code must retain the above copyright
+ *         notice, this list of conditions and the following disclaimer.
+ *       * Redistributions in binary form must reproduce the above copyright
+ *         notice, this list of conditions and the following disclaimer in the
+ *         documentation and/or other materials provided with the distribution.
+ *
+ * THIS SOFTWARE IS PROVIDED ''AS IS'' AND ANY
+ * EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
+ * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
+ * DISCLAIMED. IN NO EVENT SHALL Rambler BE LIABLE FOR ANY
+ * DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
+ * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
+ * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
+ * ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+ * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ */
+
+
+#ifndef DIFF_H_
+#define DIFF_H_
+
+#include "config.h"
+#include "fstring.h"
+
+typedef enum
+{
+	DIFF_MATCH = 1,
+	DIFF_DELETE,
+	DIFF_INSERT
+} diff_op;
+
+struct diff_edit
+{
+	gshort op;
+	gint off; /* off ginto s1 if MATCH or DELETE but s2 if INSERT */
+	gint len;
+};
+
+gint rspamd_diff(const void *a, gint aoff, gint n, const void *b, gint boff, gint m,
+		gint dmax, GArray *ses, gint *sn);
+guint32 compare_diff_distance (f_str_t *s1, f_str_t *s2);
+
+#endif /* DIFF_H_ */
diff --git a/src/expressions.c b/src/expressions.c
index c2af8288c..3dfd542a4 100644
--- a/src/expressions.c
+++ b/src/expressions.c
@@ -31,6 +31,7 @@
 #include "expressions.h"
 #include "html.h"
 #include "lua/lua_common.h"
+#include "diff.h"
 
 gboolean                        rspamd_compare_encoding (struct worker_task *task, GList * args, void *unused);
 gboolean                        rspamd_header_exists (struct worker_task *task, GList * args, void *unused);
@@ -1083,7 +1084,12 @@ rspamd_parts_distance (struct worker_task * task, GList * args, void *unused)
 			return FALSE;
 		}
 		if (!p1->is_empty && !p2->is_empty) {
-			diff = fuzzy_compare_parts (p1, p2);
+			if (p1->diff_str != NULL && p2->diff_str != NULL) {
+				diff = compare_diff_distance (p1->diff_str, p2->diff_str);
+			}
+			else {
+				diff = fuzzy_compare_parts (p1, p2);
+			}
 			debug_task ("got likeliness between parts of %d%%, threshold is %d%%", diff, threshold);
 			*pdiff = diff;
 			memory_pool_set_variable (task->task_pool, "parts_distance", pdiff, NULL);
diff --git a/src/fstring.c b/src/fstring.c
index 84c8c54bd..3f7ae2195 100644
--- a/src/fstring.c
+++ b/src/fstring.c
@@ -241,6 +241,23 @@ fstrpush (f_str_t * dest, gchar c)
 }
 
 /*
+ * Push one character to fstr
+ */
+gint
+fstrpush_unichar (f_str_t * dest, gunichar c)
+{
+	int                            l;
+	if (dest->size < dest->len) {
+		/* Need to reallocate string */
+		return 0;
+	}
+
+	l = g_unichar_to_utf8 (c, dest->begin + dest->len);
+	dest->len += l;
+	return l;
+}
+
+/*
  * Allocate memory for f_str_t
  */
 f_str_t                        *
diff --git a/src/fstring.h b/src/fstring.h
index eb32dc285..566cbdc32 100644
--- a/src/fstring.h
+++ b/src/fstring.h
@@ -69,6 +69,11 @@ size_t fstrcat (f_str_t *dest, f_str_t *src);
 gint fstrpush (f_str_t *dest, gchar c);
 
 /*
+ * Push one character to fstr
+ */
+gint fstrpush_unichar (f_str_t *dest, gunichar c);
+
+/*
  * Allocate memory for f_str_t
  */
 f_str_t* fstralloc (memory_pool_t *pool, size_t len);
diff --git a/src/fuzzy.c b/src/fuzzy.c
index 5a2decb34..2639d68a7 100644
--- a/src/fuzzy.c
+++ b/src/fuzzy.c
@@ -313,7 +313,7 @@ fuzzy_init_byte_array (GByteArray * in, memory_pool_t * pool)
 }
 
 void
-fuzzy_init_part (struct mime_text_part *part, memory_pool_t *pool)
+fuzzy_init_part (struct mime_text_part *part, memory_pool_t *pool, gsize max_diff)
 {
 	fuzzy_hash_t                   *new, *new2;
 	gchar                          *c, *end, *begin;
@@ -321,7 +321,7 @@ fuzzy_init_part (struct mime_text_part *part, memory_pool_t *pool)
 	GList                          *cur_offset;
 	struct process_exception       *cur_ex = NULL;
 	gunichar                        uc;
-	GString *debug;
+	gboolean                        write_diff = FALSE;
 
 	cur_offset = part->urls_offset;
 	if (cur_offset != NULL) {
@@ -371,7 +371,15 @@ fuzzy_init_part (struct mime_text_part *part, memory_pool_t *pool)
 		}
 	}
 
-	debug = g_string_sized_new (real_len);
+	write_diff = real_len < max_diff;
+
+	if (write_diff) {
+		part->diff_str = fstralloc (pool, real_len);
+	}
+	else {
+		part->diff_str = NULL;
+	}
+
 	new->block_size = fuzzy_blocksize (real_len);
 	new2->block_size = new->block_size * 2;
 
@@ -397,7 +405,9 @@ fuzzy_init_part (struct mime_text_part *part, memory_pool_t *pool)
 				uc = g_utf8_get_char (c);
 				if (g_unichar_isalnum (uc)) {
 					fuzzy_update2 (new, new2, uc);
-					g_string_append_unichar (debug, uc);
+					if (write_diff) {
+						fstrpush_unichar (part->diff_str, uc);
+					}
 				}
 				c = g_utf8_next_char (c);
 			}
@@ -415,13 +425,15 @@ fuzzy_init_part (struct mime_text_part *part, memory_pool_t *pool)
 			else {
 				if (!g_ascii_isspace (*c) && !g_ascii_ispunct (*c)) {
 					fuzzy_update2 (new, new2, *c);
-					g_string_append_c (debug, *c);
+					if (write_diff) {
+						fstrpush (part->diff_str, *c);
+					}
 				}
 				c++;
 			}
 		}
 	}
-	msg_info ("make hash of string: %v", debug);
+
 	/* Check whether we have more bytes in a rolling window */
 	if (new->rh != 0) {
 		new->hash_pipe[new->hi] = b64[new->h % 64];
diff --git a/src/fuzzy.h b/src/fuzzy.h
index 271bfee2a..a1daa5107 100644
--- a/src/fuzzy.h
+++ b/src/fuzzy.h
@@ -30,7 +30,7 @@ struct mime_text_part;
  */
 fuzzy_hash_t * fuzzy_init (f_str_t *in, memory_pool_t *pool);
 fuzzy_hash_t * fuzzy_init_byte_array (GByteArray *in, memory_pool_t *pool);
-void fuzzy_init_part (struct mime_text_part *part, memory_pool_t *pool);
+void fuzzy_init_part (struct mime_text_part *part, memory_pool_t *pool, gsize max_diff);
 
 gint fuzzy_compare_parts (struct mime_text_part *p1, struct mime_text_part *p2);
 
diff --git a/src/lua/lua_task.c b/src/lua/lua_task.c
index 4916a50ad..f217ee0b2 100644
--- a/src/lua/lua_task.c
+++ b/src/lua/lua_task.c
@@ -37,6 +37,7 @@
 #include "../classifiers/classifiers.h"
 #include "../binlog.h"
 #include "../statfile_sync.h"
+#include "../diff.h"
 
 extern stat_file_t* get_statfile_by_symbol (statfile_pool_t *pool, struct classifier_config *ccf,
 		const gchar *symbol, struct statfile **st, gboolean try_create);
@@ -1368,7 +1369,12 @@ lua_textpart_compare_distance (lua_State * L)
 		}
 		else {
 			if (!part->is_empty && !other->is_empty) {
-				diff = fuzzy_compare_parts (part, other);
+				if (part->diff_str != NULL && other->diff_str != NULL) {
+					diff = compare_diff_distance (part->diff_str, other->diff_str);
+				}
+				else {
+					diff = fuzzy_compare_parts (part, other);
+				}
 			}
 			else if ((part->is_empty && !other->is_empty) || (!part->is_empty && other->is_empty)) {
 				/* Empty and non empty parts are different */
diff --git a/src/message.c b/src/message.c
index 8ff53ea93..93c9fb622 100644
--- a/src/message.c
+++ b/src/message.c
@@ -786,7 +786,7 @@ process_text_part (struct worker_task *task, GByteArray *part_content, GMimeCont
 #endif
 		}
 
-		fuzzy_init_part (text_part, task->task_pool);
+		fuzzy_init_part (text_part, task->task_pool, task->cfg->max_diff);
 		memory_pool_add_destructor (task->task_pool, (pool_destruct_func) free_byte_array_callback, text_part->content);
 		task->text_parts = g_list_prepend (task->text_parts, text_part);
 	}
@@ -806,7 +806,7 @@ process_text_part (struct worker_task *task, GByteArray *part_content, GMimeCont
 		text_part->orig = convert_text_to_utf (task, part_content, type, text_part);
 		text_part->content = text_part->orig;
 		url_parse_text (task->task_pool, task, text_part, FALSE);
-		fuzzy_init_part (text_part, task->task_pool);
+		fuzzy_init_part (text_part, task->task_pool, task->cfg->max_diff);
 		task->text_parts = g_list_prepend (task->text_parts, text_part);
 	}
 }
diff --git a/src/message.h b/src/message.h
index 15fd90188..d716f7906 100644
--- a/src/message.h
+++ b/src/message.h
@@ -35,6 +35,7 @@ struct mime_text_part {
 	fuzzy_hash_t *double_fuzzy;
 	GMimeObject *parent;
 	GUnicodeScript script;
+	f_str_t *diff_str;
 };
 
 struct received_header {