Reduce memory allocated by var_export, json_encode, serialize, and other (#8902)

smart_str uses an over-allocated string to optimize for append operations. Functions that use smart_str tend to return the over-allocated string directly. This results in unnecessary memory usage, especially for small strings.

The overhead can be up to 231 bytes for strings smaller than that, and 4095 for other strings. This can be avoided for strings smaller than `4096 - zend_string header size - 1` by reallocating the string.

This change introduces `smart_str_trim_to_size()`, and calls it in `smart_str_extract()`. Functions that use `smart_str` are updated to use `smart_str_extract()`.

Fixes GH-8896

NEWS

@@ -2,11 +2,16 @@ PHP                                                                        NEWS
 |||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
 ?? ??? ????, PHP 8.2.0beta1
 
+- Core:
+  . Reduced the memory footprint of strings returned by var_export(),
+    json_encode(), serialize(), iconv_*(), mb_ereg*(), session_create_id(),
+    http_build_query(), strstr(), Reflection*::__toString(). (Arnaud)
+
 - CLI:
-   - Updated the mime-type table for the builtin-server. (Ayesh Karunaratne)
+  . Updated the mime-type table for the builtin-server. (Ayesh Karunaratne)
 
 - FPM:
-  - Added listen.setfib pool option to set route FIB on FreeBSD. (David Carlier)
+  . Added listen.setfib pool option to set route FIB on FreeBSD. (David Carlier)
 
 07 Jul 2022, PHP 8.2.0alpha3
 

UPGRADING.INTERNALS

@@ -39,6 +39,13 @@ PHP 8.2 INTERNALS UPGRADE NOTES
   - zend_object_do_operation_t
 * Added a new zero_position argument to php_stream_fopen_from_fd_rel to reflect
   if this a newly created file so the current file offset needs not to be checked.
+* Added smart_str_trim_to_size(). The function trims the memory allocated for the
+  string. This can considerably reduce the memory footprint of strings smaller
+  than approximately 4096 bytes.
+* smart_str_extract() and the spprintf family of functions now use
+  smart_str_trim_to_size() before returning the string.
+* It is recommended to use smart_str_extract() or smart_str_trim_to_size() when
+  using the smart_str API.
 
 ========================
 2. Build system changes

Zend/zend.c

@@ -272,8 +272,7 @@ ZEND_API zend_string *zend_vstrpprintf(size_t max_len, const char *format, va_li
 		ZSTR_LEN(buf.s) = max_len;
 	}
 
-	smart_str_0(&buf);
-	return buf.s;
+	return smart_str_extract(&buf);
 }
 /* }}} */
 

Zend/zend_smart_str.h

@@ -25,6 +25,10 @@
 	smart_str_appendl_ex((dest), (src), strlen(src), (what))
 #define smart_str_appends(dest, src) \
 	smart_str_appendl((dest), (src), strlen(src))
+#define smart_str_extract(dest) \
+	smart_str_extract_ex((dest), 0)
+#define smart_str_trim_to_size(dest) \
+	smart_str_trim_to_size_ex((dest), 0)
 #define smart_str_extend(dest, len) \
 	smart_str_extend_ex((dest), (len), 0)
 #define smart_str_appendc(dest, c) \
@@ -101,10 +105,19 @@ static zend_always_inline size_t smart_str_get_len(smart_str *str) {
 	return str->s ? ZSTR_LEN(str->s) : 0;
 }
 
-static zend_always_inline zend_string *smart_str_extract(smart_str *str) {
+static zend_always_inline void smart_str_trim_to_size_ex(smart_str *str, bool persistent)
+{
+	if (str->s && str->a > ZSTR_LEN(str->s)) {
+		str->s = zend_string_realloc(str->s, ZSTR_LEN(str->s), persistent);
+		str->a = ZSTR_LEN(str->s);
+	}
+}
+
+static zend_always_inline zend_string *smart_str_extract_ex(smart_str *str, bool persistent) {
 	if (str->s) {
 		zend_string *res;
 		smart_str_0(str);
+		smart_str_trim_to_size_ex(str, persistent);
 		res = str->s;
 		str->s = NULL;
 		return res;

Zend/zend_smart_str_public.h

@@ -18,6 +18,7 @@
 #define ZEND_SMART_STR_PUBLIC_H
 
 typedef struct {
+	/** See smart_str_extract() */
 	zend_string *s;
 	size_t a;
 } smart_str;

ext/dom/documenttype.c

@@ -188,8 +188,7 @@ int dom_documenttype_internal_subset_read(dom_object *obj, zval *retval)
 		}
 
 		if (ret_buf.s) {
-			smart_str_0(&ret_buf);
-			ZVAL_NEW_STR(retval, ret_buf.s);
+			ZVAL_STR(retval, smart_str_extract(&ret_buf));
 			return SUCCESS;
 		}
 	}

ext/filter/sanitizing_filters.c

@@ -46,9 +46,8 @@ static void php_filter_encode_html(zval *value, const unsigned char *chars)
 		s++;
 	}
 
-	smart_str_0(&str);
 	zval_ptr_dtor(value);
-	ZVAL_NEW_STR(value, str.s);
+	ZVAL_STR(value, smart_str_extract(&str));
 }
 
 static const unsigned char hexchars[] = "0123456789ABCDEF";

ext/iconv/iconv.c

@@ -1845,7 +1845,7 @@ PHP_FUNCTION(iconv_substr)
 	_php_iconv_show_error(err, GENERIC_SUPERSET_NAME, charset);
 
 	if (err == PHP_ICONV_ERR_SUCCESS && retval.s != NULL) {
-		RETURN_NEW_STR(retval.s);
+		RETURN_STR(smart_str_extract(&retval));
 	}
 	smart_str_free(&retval);
 	RETURN_FALSE;
@@ -2038,11 +2038,7 @@ PHP_FUNCTION(iconv_mime_encode)
 	_php_iconv_show_error(err, out_charset, in_charset);
 
 	if (err == PHP_ICONV_ERR_SUCCESS) {
-		if (retval.s != NULL) {
-			RETVAL_STR(retval.s);
-		} else {
-			RETVAL_EMPTY_STRING();
-		}
+		RETVAL_STR(smart_str_extract(&retval));
 	} else {
 		smart_str_free(&retval);
 		RETVAL_FALSE;
@@ -2083,11 +2079,7 @@ PHP_FUNCTION(iconv_mime_decode)
 	_php_iconv_show_error(err, charset, "???");
 
 	if (err == PHP_ICONV_ERR_SUCCESS) {
-		if (retval.s != NULL) {
-			RETVAL_STR(retval.s);
-		} else {
-			RETVAL_EMPTY_STRING();
-		}
+		RETVAL_STR(smart_str_extract(&retval));
 	} else {
 		smart_str_free(&retval);
 		RETVAL_FALSE;

ext/json/json.c

@@ -212,11 +212,7 @@ PHP_FUNCTION(json_encode)
 		}
 	}
 
-	smart_str_0(&buf); /* copy? */
-	if (buf.s) {
-		RETURN_NEW_STR(buf.s);
-	}
-	RETURN_EMPTY_STRING();
+	RETURN_STR(smart_str_extract(&buf));
 }
 /* }}} */
 

ext/mbstring/php_mbregex.c

@@ -1144,13 +1144,10 @@ static void _php_mb_regex_ereg_replace_exec(INTERNAL_FUNCTION_PARAMETERS, OnigOp
 
 	if (err <= -2) {
 		smart_str_free(&out_buf);
-		RETVAL_FALSE;
-	} else if (out_buf.s) {
-		smart_str_0(&out_buf);
-		RETVAL_STR(out_buf.s);
-	} else {
-		RETVAL_EMPTY_STRING();
+		RETURN_FALSE;
 	}
+
+	RETURN_STR(smart_str_extract(&out_buf));
 }
 /* }}} */
 

ext/session/session.c

@@ -1464,7 +1464,7 @@ PHPAPI zend_result php_session_reset_id(void) /* {{{ */
 		smart_str_0(&var);
 		if (sid) {
 			zval_ptr_dtor_str(sid);
-			ZVAL_NEW_STR(sid, var.s);
+			ZVAL_STR(sid, smart_str_extract(&var));
 		} else {
 			REGISTER_STRINGL_CONSTANT("SID", ZSTR_VAL(var.s), ZSTR_LEN(var.s), 0);
 			smart_str_free(&var);
@@ -2362,8 +2362,7 @@ PHP_FUNCTION(session_create_id)
 		php_error_docref(NULL, E_WARNING, "Failed to create new ID");
 		RETURN_FALSE;
 	}
-	smart_str_0(&id);
-	RETVAL_NEW_STR(id.s);
+	RETVAL_STR(smart_str_extract(&id));
 }
 /* }}} */
 

ext/soap/php_sdl.c

@@ -3261,8 +3261,7 @@ sdlPtr get_sdl(zval *this_ptr, char *uri, zend_long cache_wsdl)
 			smart_str_appends(&proxy,Z_STRVAL_P(proxy_host));
 			smart_str_appends(&proxy,":");
 			smart_str_append_long(&proxy,Z_LVAL_P(proxy_port));
-			smart_str_0(&proxy);
-			ZVAL_NEW_STR(&str_proxy, proxy.s);
+			ZVAL_STR(&str_proxy, smart_str_extract(&proxy));
 
 			if (!context) {
 				context = php_stream_context_alloc();
@@ -3304,8 +3303,7 @@ sdlPtr get_sdl(zval *this_ptr, char *uri, zend_long cache_wsdl)
 			http_context_headers(context, has_authorization, has_proxy_authorization, 0, &headers);
 		}
 
-		smart_str_0(&headers);
-		ZVAL_NEW_STR(&str_headers, headers.s);
+		ZVAL_STR(&str_headers, smart_str_extract(&headers));
 		php_stream_context_set_option(context, "http", "header", &str_headers);
 		zval_ptr_dtor(&str_headers);
 	}

ext/spl/spl_array.c

@@ -1554,7 +1554,7 @@ PHP_METHOD(ArrayObject, serialize)
 	/* done */
 	PHP_VAR_SERIALIZE_DESTROY(var_hash);
 
-	RETURN_NEW_STR(buf.s);
+	RETURN_STR(smart_str_extract(&buf));
 } /* }}} */
 
 /* {{{ unserialize the object */

ext/spl/spl_dllist.c

@@ -1032,12 +1032,10 @@ PHP_METHOD(SplDoublyLinkedList, serialize)
 		current = next;
 	}
 
-	smart_str_0(&buf);
-
 	/* done */
 	PHP_VAR_SERIALIZE_DESTROY(var_hash);
 
-	RETURN_NEW_STR(buf.s);
+	RETURN_STR(smart_str_extract(&buf));
 } /* }}} */
 
 /* {{{ Unserializes storage */

ext/spl/spl_observer.c

@@ -808,7 +808,7 @@ PHP_METHOD(SplObjectStorage, serialize)
 	/* done */
 	PHP_VAR_SERIALIZE_DESTROY(var_hash);
 
-	RETURN_NEW_STR(buf.s);
+	RETURN_STR(smart_str_extract(&buf));
 } /* }}} */
 
 /* {{{ Unserializes storage */

ext/standard/http.c

@@ -238,12 +238,6 @@ PHP_FUNCTION(http_build_query)
 
 	php_url_encode_hash_ex(HASH_OF(formdata), &formstr, prefix, prefix_len, NULL, 0, NULL, 0, (Z_TYPE_P(formdata) == IS_OBJECT ? formdata : NULL), arg_sep, (int)enc_type);
 
-	if (!formstr.s) {
-		RETURN_EMPTY_STRING();
-	}
-
-	smart_str_0(&formstr);
-
-	RETURN_NEW_STR(formstr.s);
+	RETURN_STR(smart_str_extract(&formstr));
 }
 /* }}} */

ext/standard/string.c

@@ -2891,8 +2891,7 @@ static void php_strtr_array(zval *return_value, zend_string *input, HashTable *p
 
 	if (result.s) {
 		smart_str_appendl(&result, str + old_pos, slen - old_pos);
-		smart_str_0(&result);
-		RETVAL_NEW_STR(result.s);
+		RETVAL_STR(smart_str_extract(&result));
 	} else {
 		smart_str_free(&result);
 		RETVAL_STR_COPY(input);

ext/standard/var.c

@@ -645,7 +645,7 @@ PHP_FUNCTION(var_export)
 	smart_str_0 (&buf);
 
 	if (return_output) {
-		RETURN_NEW_STR(buf.s);
+		RETURN_STR(smart_str_extract(&buf));
 	} else {
 		PHPWRITE(ZSTR_VAL(buf.s), ZSTR_LEN(buf.s));
 		smart_str_free(&buf);
@@ -1318,11 +1318,7 @@ PHP_FUNCTION(serialize)
 		RETURN_THROWS();
 	}
 
-	if (buf.s) {
-		RETURN_NEW_STR(buf.s);
-	} else {
-		RETURN_EMPTY_STRING();
-	}
+	RETURN_STR(smart_str_extract(&buf));
 }
 /* }}} */