zend_compile: Optimize sprintf() into a rope

TimWolla · TimWolla · commit 9fc7a7328c42 · 2024-06-12T11:49:19.000+02:00
This optimization will compile `sprintf()` using only `%s` placeholders into a
rope at compile time, effectively making those calls equivalent to the use of
string interpolation, with the added benefit of supporting arbitrary
expressions instead of just expressions starting with a `$`.

For a synthetic test using:

    &lt;?php

    $a = 'foo';
    $b = 'bar';

    for ($i = 0; $i &lt; 100_000_000; $i++) {
    	sprintf("%s-%s", $a, $b);
    }

This optimization yields a 2.1× performance improvement:

    $ hyperfine 'sapi/cli/php -d zend_extension=php-src/modules/opcache.so -d opcache.enable_cli=1 test.php' \
          '/tmp/unoptimized -d zend_extension=php-src/modules/opcache.so -d opcache.enable_cli=1 test.php'
    Benchmark 1: sapi/cli/php -d zend_extension=php-src/modules/opcache.so -d opcache.enable_cli=1 test.php
      Time (mean ± σ):      1.869 s ±  0.033 s    [User: 1.865 s, System: 0.003 s]
      Range (min … max):    1.840 s …  1.945 s    10 runs

    Benchmark 2: /tmp/unoptimized -d zend_extension=php-src/modules/opcache.so -d opcache.enable_cli=1 test.php
      Time (mean ± σ):      4.011 s ±  0.034 s    [User: 4.006 s, System: 0.005 s]
      Range (min … max):    3.964 s …  4.079 s    10 runs

    Summary
      sapi/cli/php -d zend_extension=php-src/modules/opcache.so -d opcache.enable_cli=1 test.php ran
        2.15 ± 0.04 times faster than /tmp/unoptimized -d zend_extension=php-src/modules/opcache.so -d opcache.enable_cli=1 test.php

This optimization comes with a small and probably insignificant behavioral
change: If one of the values cannot be (cleanly) converted to a string, for
example when attempting to insert an object that is not `Stringable`, the
resulting Exception will naturally not show the `sprintf()` call in the
resulting stack trace, because there is no call to `sprintf()`.

Nevertheless it will correctly point out the line of the `sprintf()` call as
the source of the Exception, pointing the user towards the correct location.
diff --git a/Zend/zend_compile.c b/Zend/zend_compile.c
@@ -4712,6 +4712,171 @@ static void zend_compile_ns_call(znode *result, znode *name_node, zend_ast *args
 }
 /* }}} */
 
+static zend_op *zend_compile_rope_add(znode *result, uint32_t num, znode *elem_node);
+static zend_op *zend_compile_rope_add_ex(zend_op *opline, znode *result, uint32_t num, znode *elem_node);
+static void zend_compile_rope_finalize(znode *result, uint32_t j, zend_op *init_opline, zend_op *opline);
+
+static zend_result zend_compile_func_sprintf(znode *result, zend_ast_list *args) /* {{{ */
+{
+	if (args->children < 1) {
+		return FAILURE;
+	}
+
+	zend_eval_const_expr(&args->child[0]);
+	if (args->child[0]->kind != ZEND_AST_ZVAL) {
+		return FAILURE;
+	}
+
+	zval *format_string = zend_ast_get_zval(args->child[0]);
+	if (Z_TYPE_P(format_string) != IS_STRING) {
+		return FAILURE;
+	}
+	if (Z_STRLEN_P(format_string) >= 256) {
+		return FAILURE;
+	}
+
+	char *p;
+	char *end;
+	uint32_t string_placeholder_count;
+	
+	string_placeholder_count = 0;
+	p = Z_STRVAL_P(format_string);
+	end = p + Z_STRLEN_P(format_string);
+
+	for (;;) {
+		p = memchr(p, '%', end - p);
+		if (!p) {
+			break;
+		}
+
+		char *q = p + 1;
+		if (q == end) {
+			return FAILURE;
+		}
+
+		switch (*q) {
+		case 's':
+			string_placeholder_count++;
+			break;
+		case '%':
+			break;
+		default:
+			return FAILURE;
+		}
+
+		p = q;
+		p++;
+	}
+
+	/* Bail out if the number of placeholders does not match the number of values. */
+	if (string_placeholder_count != (args->children - 1)) {
+		return FAILURE;
+	}
+
+	znode *elements = NULL;
+
+	if (string_placeholder_count > 0) {
+		elements = safe_emalloc(sizeof(*elements), string_placeholder_count, 0);
+	}
+
+	/* Compile the value expressions first for error handling that is consistent
+	 * with a function call: Values that fail to convert to a string may emit errors.
+	 */
+	for (uint32_t i = 0; i < string_placeholder_count; i++) {
+		zend_compile_expr(elements + i, args->child[1 + i]);
+		if (elements[i].op_type == IS_CONST) {
+			if (Z_TYPE(elements[i].u.constant) != IS_ARRAY) {
+				convert_to_string(&elements[i].u.constant);
+			}
+		}
+	}
+
+	uint32_t rope_elements = 0;
+	uint32_t rope_init_lineno = -1;
+	zend_op *opline = NULL;
+
+	string_placeholder_count = 0;
+	p = Z_STRVAL_P(format_string);
+	end = p + Z_STRLEN_P(format_string);
+	char *offset = p;
+	for (;;) {
+		p = memchr(p, '%', end - p);
+		if (!p) {
+			break;
+		}
+
+		char *q = p + 1;
+		ZEND_ASSERT(q < end);
+		ZEND_ASSERT(*q == 's' || *q == '%');
+
+		if (*q == '%') {
+			/* Optimization to not create a dedicated rope element for the literal '%':
+			 * Include the first '%' within the "constant" part instead of dropping the
+			 * full placeholder.
+			 */
+			p++;
+		}
+
+		if (p != offset) {
+			znode const_node;
+			const_node.op_type = IS_CONST;
+			ZVAL_STRINGL(&const_node.u.constant, offset, p - offset);
+			if (rope_elements == 0) {
+				rope_init_lineno = get_next_op_number();
+			}
+			opline = zend_compile_rope_add(result, rope_elements++, &const_node);
+		}
+
+		if (*q == 's') {
+			/* Perform the cast of constant arrays when actually evaluating corresponding placeholder
+			 * for correct error reporting.
+			 */
+			if (elements[string_placeholder_count].op_type == IS_CONST) {
+				if (Z_TYPE(elements[string_placeholder_count].u.constant) == IS_ARRAY) {
+					zend_emit_op_tmp(&elements[string_placeholder_count], ZEND_CAST, &elements[string_placeholder_count], NULL)->extended_value = IS_STRING;
+				}
+			}
+			if (rope_elements == 0) {
+				rope_init_lineno = get_next_op_number();
+			}
+			opline = zend_compile_rope_add(result, rope_elements++, &elements[string_placeholder_count]);
+
+			string_placeholder_count++;
+		}
+
+		p = q;
+		p++;
+		offset = p;
+	}
+	if (end != offset) {
+		/* Add the constant part after the last placeholder. */
+		znode const_node;
+		const_node.op_type = IS_CONST;
+		ZVAL_STRINGL(&const_node.u.constant, offset, end - offset);
+		if (rope_elements == 0) {
+			rope_init_lineno = get_next_op_number();
+		}
+		opline = zend_compile_rope_add(result, rope_elements++, &const_node);
+	}
+	if (rope_elements == 0) {
+		/* Handle empty format strings. */
+		znode const_node;
+		const_node.op_type = IS_CONST;
+		ZVAL_EMPTY_STRING(&const_node.u.constant);
+		if (rope_elements == 0) {
+			rope_init_lineno = get_next_op_number();
+		}
+		opline = zend_compile_rope_add(result, rope_elements++, &const_node);
+	}
+	ZEND_ASSERT(opline != NULL);
+
+	zend_op *init_opline = CG(active_op_array)->opcodes + rope_init_lineno;
+	zend_compile_rope_finalize(result, rope_elements, init_opline, opline);
+	efree(elements);
+
+	return SUCCESS;
+}
+
 static zend_result zend_try_compile_special_func_ex(znode *result, zend_string *lcname, zend_ast_list *args, zend_function *fbc, uint32_t type) /* {{{ */
 {
 	if (zend_string_equals_literal(lcname, "strlen")) {
@@ -4778,6 +4943,8 @@ static zend_result zend_try_compile_special_func_ex(znode *result, zend_string *
 		return zend_compile_func_array_slice(result, args);
 	} else if (zend_string_equals_literal(lcname, "array_key_exists")) {
 		return zend_compile_func_array_key_exists(result, args);
+	} else if (zend_string_equals_literal(lcname, "sprintf")) {
+		return zend_compile_func_sprintf(result, args);	
 	} else {
 		return FAILURE;
 	}
diff --git a/ext/standard/tests/strings/sprintf_rope_optimization_001.phpt b/ext/standard/tests/strings/sprintf_rope_optimization_001.phpt
@@ -0,0 +1,167 @@
+--TEST--
+Test sprintf() function : Rope Optimization
+--FILE--
+<?php
+function func($str) {
+	return strtoupper($str);
+}
+function sideeffect() {
+	echo "Called!\n";
+	return "foo";
+}
+class Foo {
+	public function __construct() {
+		echo "Called\n";
+	}
+}
+
+$a = "foo";
+$b = "bar";
+$c = new stdClass();
+
+try {
+	var_dump(sprintf("const"));
+} catch (\Throwable $e) {echo $e, PHP_EOL; } echo PHP_EOL;
+
+try {
+	var_dump(sprintf("%s", $a));
+} catch (\Throwable $e) {echo $e, PHP_EOL; } echo PHP_EOL;
+
+try {
+	var_dump(sprintf("%s/%s", $a, $b));
+} catch (\Throwable $e) {echo $e, PHP_EOL; } echo PHP_EOL;
+
+try {
+	var_dump(sprintf("%s/%s/%s", $a, $b));
+} catch (\Throwable $e) {echo $e, PHP_EOL; } echo PHP_EOL;
+
+try {
+	var_dump(sprintf("%s/%s/%s", $a, $b, $c));
+} catch (\Throwable $e) {echo $e, PHP_EOL; } echo PHP_EOL;
+
+try {
+	var_dump(sprintf("%s/", func("baz")));
+} catch (\Throwable $e) {echo $e, PHP_EOL; } echo PHP_EOL;
+
+try {
+	var_dump(sprintf("/%s", func("baz")));
+} catch (\Throwable $e) {echo $e, PHP_EOL; } echo PHP_EOL;
+
+try {
+	var_dump(sprintf("/%s/", func("baz")));
+} catch (\Throwable $e) {echo $e, PHP_EOL; } echo PHP_EOL;
+
+try {
+	var_dump(sprintf("%s%s%s%s", $a, $b, func("baz"), $a));
+} catch (\Throwable $e) {echo $e, PHP_EOL; } echo PHP_EOL;
+
+try {
+	var_dump(sprintf("%s/%s", sprintf("%s:%s", $a, $b), sprintf("%s-%s", func('baz'), func('baz'))));
+} catch (\Throwable $e) {echo $e, PHP_EOL; } echo PHP_EOL;
+
+try {
+	var_dump(sprintf(sideeffect()));
+} catch (\Throwable $e) {echo $e, PHP_EOL; } echo PHP_EOL;
+
+try {
+	var_dump(sprintf("%s-%s-%s", __FILE__, __LINE__, 1));
+} catch (\Throwable $e) {echo $e, PHP_EOL; } echo PHP_EOL;
+
+try {
+	$values = range('a', 'z');
+	var_dump(sprintf("%s%s%s", "{$values[0]}{$values[1]}{$values[2]}", "{$values[3]}{$values[4]}{$values[5]}", "{$values[6]}{$values[7]}{$values[8]}"));
+} catch (\Throwable $e) {echo $e, PHP_EOL; } echo PHP_EOL;
+
+try {
+	var_dump(sprintf("%s%s%s", new Foo(), new Foo(), new Foo(), ));
+} catch (\Throwable $e) {echo $e, PHP_EOL; } echo PHP_EOL;
+
+try {
+	var_dump(sprintf(...));
+} catch (\Throwable $e) {echo $e, PHP_EOL; } echo PHP_EOL;
+
+try {
+	var_dump(sprintf('%%s'));
+} catch (\Throwable $e) {echo $e, PHP_EOL; } echo PHP_EOL;
+
+try {
+	var_dump(sprintf('%%s', 'test'));
+} catch (\Throwable $e) {echo $e, PHP_EOL; } echo PHP_EOL;
+
+try {
+	var_dump(sprintf('%s-%s-%s', [], [], []));
+} catch (\Throwable $e) {echo $e, PHP_EOL; } echo PHP_EOL;
+
+try {
+	var_dump(sprintf(""));
+} catch (\Throwable $e) {echo $e, PHP_EOL; } echo PHP_EOL;
+
+echo "Done";
+?>
+--EXPECTF--
+string(5) "const"
+
+string(3) "foo"
+
+string(7) "foo/bar"
+
+ArgumentCountError: 4 arguments are required, 3 given in %s:32
+Stack trace:
+#0 %s(32): sprintf('%s/%s/%s', 'foo', 'bar')
+#1 {main}
+
+Error: Object of class stdClass could not be converted to string in %s:36
+Stack trace:
+#0 {main}
+
+string(4) "BAZ/"
+
+string(4) "/BAZ"
+
+string(5) "/BAZ/"
+
+string(12) "foobarBAZfoo"
+
+string(15) "foo:bar/BAZ-BAZ"
+
+Called!
+string(3) "foo"
+
+string(%d) "%ssprintf_rope_optimization_001.php-%d-1"
+
+string(9) "abcdefghi"
+
+Called
+Called
+Called
+Error: Object of class Foo could not be converted to string in %s:73
+Stack trace:
+#0 {main}
+
+object(Closure)#3 (2) {
+  ["function"]=>
+  string(7) "sprintf"
+  ["parameter"]=>
+  array(2) {
+    ["$format"]=>
+    string(10) "<required>"
+    ["$values"]=>
+    string(10) "<optional>"
+  }
+}
+
+string(2) "%s"
+
+string(2) "%s"
+
+
+Warning: Array to string conversion in %s on line 89
+
+Warning: Array to string conversion in %s on line 89
+
+Warning: Array to string conversion in %s on line 89
+string(17) "Array-Array-Array"
+
+string(0) ""
+
+Done
diff --git a/ext/standard/tests/strings/sprintf_rope_optimization_002.phpt b/ext/standard/tests/strings/sprintf_rope_optimization_002.phpt
@@ -0,0 +1,26 @@
+--TEST--
+Test sprintf() function : Rope Optimization with a throwing error handler.
+--FILE--
+<?php
+
+function exception_error_handler(int $errno, string $errstr, ?string $errfile, int $errline) {
+	if (!(error_reporting() & $errno)) {
+		// This error code is not included in error_reporting
+		return;
+	}
+	throw new \ErrorException($errstr, 0, $errno, $errfile, $errline);
+}
+set_error_handler(exception_error_handler(...));
+
+try {
+	var_dump(sprintf("%s-%s", new stdClass(), []));
+} catch (\Throwable $e) {echo $e, PHP_EOL; } echo PHP_EOL;
+
+echo "Done";
+?>
+--EXPECTF--
+Error: Object of class stdClass could not be converted to string in %s:13
+Stack trace:
+#0 {main}
+
+Done
