Re: [Tarantool-patches] [PATCH v1 8/8] sql: refactor LIKE() function

[Mergen Imeev via Tarantool-patches <tarantool-patches@dev.tarantool.org>]

Thank you for the review! My answer, diff and new patch below. Also, I replaced
self-created and SQLite3 functions to work with UTF8 characters by
iucnv_getNextUChar(). And I replaced sql_result_bool() by mem_set_bool(), so I
will be able to drop sql_result_bool() later.

On Sat, Oct 09, 2021 at 12:02:35AM +0200, Vladislav Shpilevoy wrote:
> Thanks for the patch!
> 
> > diff --git a/src/box/sql/func.c b/src/box/sql/func.c
> > index 1b4d52225..27106ced6 100644
> > --- a/src/box/sql/func.c
> > +++ b/src/box/sql/func.c
> > @@ -1281,46 +1281,28 @@ likeFunc(sql_context *context, int argc, struct Mem *argv)
> >  {
> >  	u32 escape = SQL_END_OF_STRING;
> >  	int nPat;
> > -	if (argc != 2 && argc != 3) {
> > -		diag_set(ClientError, ER_FUNC_WRONG_ARG_COUNT,
> > -			 "LIKE", "2 or 3", argc);
> > -		context->is_aborted = true;
> > -		return;
> > -	}
> > -	sql *db = sql_context_db_handle(context);
> > -	int rhs_type = sql_value_type(&argv[0]);
> > -	int lhs_type = sql_value_type(&argv[1]);
> > -
> > -	if (lhs_type != MP_STR || rhs_type != MP_STR) {
> > -		if (lhs_type == MP_NIL || rhs_type == MP_NIL)
> > -			return;
> > -		const char *str = rhs_type != MP_STR ?
> > -				  mem_str(&argv[0]) : mem_str(&argv[1]);
> > -		diag_set(ClientError, ER_INCONSISTENT_TYPES, "string", str);
> > -		context->is_aborted = true;
> > +	assert(argc == 2 || argc == 3);
> > +	struct sql *db = sql_context_db_handle(context);
> 
> You can get the db after the is_null check below. It is not
> used when the check passes.
> 
Fixed. Dropped this line and used sql_get() instead.

> > +	if (mem_is_any_null(&argv[0], &argv[1]))
> >  		return;


Diff:

diff --git a/src/box/sql/func.c b/src/box/sql/func.c
index 90417c892..afe34f7f0 100644
--- a/src/box/sql/func.c
+++ b/src/box/sql/func.c
@@ -1261,7 +1261,6 @@ likeFunc(sql_context *context, int argc, struct Mem *argv)
 	u32 escape = SQL_END_OF_STRING;
 	int nPat;
 	assert(argc == 2 || argc == 3);
-	struct sql *db = sql_context_db_handle(context);
 	if (mem_is_any_null(&argv[0], &argv[1]))
 		return;
 	assert(mem_is_str(&argv[0]) && mem_is_str(&argv[1]));
@@ -1276,7 +1275,7 @@ likeFunc(sql_context *context, int argc, struct Mem *argv)
 	 * sql_utf8_pattern_compare().
 	 */
 	nPat = argv[0].n;
-	if (nPat > db->aLimit[SQL_LIMIT_LIKE_PATTERN_LENGTH]) {
+	if (nPat > sql_get()->aLimit[SQL_LIMIT_LIKE_PATTERN_LENGTH]) {
 		diag_set(ClientError, ER_SQL_EXECUTE, "LIKE pattern is too "\
 			 "complex");
 		context->is_aborted = true;
@@ -1284,21 +1283,23 @@ likeFunc(sql_context *context, int argc, struct Mem *argv)
 	}
 
 	if (argc == 3) {
+		if (mem_is_null(&argv[2]))
+			return;
 		/*
 		 * The escape character string must consist of a
 		 * single UTF-8 character. Otherwise, return an
 		 * error.
 		 */
-		const unsigned char *zEsc = (const unsigned char *)argv[2].z;
-		if (zEsc == 0)
-			return;
-		if (sql_utf8_char_count(zEsc, argv[2].n) != 1) {
+		UErrorCode err = U_ZERO_ERROR;
+		const char *pos = argv[2].z;
+		const char *end = argv[2].z + argv[2].n;
+		escape = ucnv_getNextUChar(icu_utf8_conv, &pos, end, &err);
+		if (pos != end || err != U_ZERO_ERROR) {
 			diag_set(ClientError, ER_SQL_EXECUTE, "ESCAPE "\
 				 "expression must be a single character");
 			context->is_aborted = true;
 			return;
 		}
-		escape = sqlUtf8Read(&zEsc);
 	}
 	if (!zA || !zB)
 		return;
@@ -1313,7 +1314,7 @@ likeFunc(sql_context *context, int argc, struct Mem *argv)
 		context->is_aborted = true;
 		return;
 	}
-	sql_result_bool(context, res == MATCH);
+	mem_set_bool(context->pOut, res == MATCH);
 }
 
 /**


New patch:

commit a739926ff3178d499650cff67c8597ff21e18b3c
Author: Mergen Imeev <imeevma@gmail.com>
Date:   Mon Sep 27 19:29:28 2021 +0300

    sql: refactor LIKE() function
    
    Part of #4145

diff --git a/src/box/sql/func.c b/src/box/sql/func.c
index 65bf03250..afe34f7f0 100644
--- a/src/box/sql/func.c
+++ b/src/box/sql/func.c
@@ -1260,63 +1260,46 @@ likeFunc(sql_context *context, int argc, struct Mem *argv)
 {
 	u32 escape = SQL_END_OF_STRING;
 	int nPat;
-	if (argc != 2 && argc != 3) {
-		diag_set(ClientError, ER_FUNC_WRONG_ARG_COUNT,
-			 "LIKE", "2 or 3", argc);
-		context->is_aborted = true;
-		return;
-	}
-	sql *db = sql_context_db_handle(context);
-	int rhs_type = sql_value_type(&argv[0]);
-	int lhs_type = sql_value_type(&argv[1]);
-
-	if (lhs_type != MP_STR || rhs_type != MP_STR) {
-		if (lhs_type == MP_NIL || rhs_type == MP_NIL)
-			return;
-		const char *str = rhs_type != MP_STR ?
-				  mem_str(&argv[0]) : mem_str(&argv[1]);
-		diag_set(ClientError, ER_INCONSISTENT_TYPES, "string", str);
-		context->is_aborted = true;
+	assert(argc == 2 || argc == 3);
+	if (mem_is_any_null(&argv[0], &argv[1]))
 		return;
-	}
-	const char *zB = mem_as_str0(&argv[0]);
-	const char *zA = mem_as_str0(&argv[1]);
-	const char *zB_end = zB + mem_len_unsafe(&argv[0]);
-	const char *zA_end = zA + mem_len_unsafe(&argv[1]);
+	assert(mem_is_str(&argv[0]) && mem_is_str(&argv[1]));
+	const char *zB = argv[0].z;
+	const char *zA = argv[1].z;
+	const char *zB_end = zB + argv[0].n;
+	const char *zA_end = zA + argv[1].n;
 
 	/*
 	 * Limit the length of the LIKE pattern to avoid problems
 	 * of deep recursion and N*N behavior in
 	 * sql_utf8_pattern_compare().
 	 */
-	nPat = mem_len_unsafe(&argv[0]);
-	testcase(nPat == db->aLimit[SQL_LIMIT_LIKE_PATTERN_LENGTH]);
-	testcase(nPat == db->aLimit[SQL_LIMIT_LIKE_PATTERN_LENGTH] + 1);
-	if (nPat > db->aLimit[SQL_LIMIT_LIKE_PATTERN_LENGTH]) {
+	nPat = argv[0].n;
+	if (nPat > sql_get()->aLimit[SQL_LIMIT_LIKE_PATTERN_LENGTH]) {
 		diag_set(ClientError, ER_SQL_EXECUTE, "LIKE pattern is too "\
 			 "complex");
 		context->is_aborted = true;
 		return;
 	}
-	/* Encoding did not change */
-	assert(zB == mem_as_str0(&argv[0]));
 
 	if (argc == 3) {
+		if (mem_is_null(&argv[2]))
+			return;
 		/*
 		 * The escape character string must consist of a
 		 * single UTF-8 character. Otherwise, return an
 		 * error.
 		 */
-		const unsigned char *zEsc = mem_as_ustr(&argv[2]);
-		if (zEsc == 0)
-			return;
-		if (sql_utf8_char_count(zEsc, mem_len_unsafe(&argv[2])) != 1) {
+		UErrorCode err = U_ZERO_ERROR;
+		const char *pos = argv[2].z;
+		const char *end = argv[2].z + argv[2].n;
+		escape = ucnv_getNextUChar(icu_utf8_conv, &pos, end, &err);
+		if (pos != end || err != U_ZERO_ERROR) {
 			diag_set(ClientError, ER_SQL_EXECUTE, "ESCAPE "\
 				 "expression must be a single character");
 			context->is_aborted = true;
 			return;
 		}
-		escape = sqlUtf8Read(&zEsc);
 	}
 	if (!zA || !zB)
 		return;
@@ -1331,7 +1314,7 @@ likeFunc(sql_context *context, int argc, struct Mem *argv)
 		context->is_aborted = true;
 		return;
 	}
-	sql_result_bool(context, res == MATCH);
+	mem_set_bool(context->pOut, res == MATCH);
 }
 
 /**