By the way, I traced the memory consumption using xdebug:

xdebug.auto_trace = on;
xdebug.show_mem_delta = on;

The first few lines are here http://pastebin.ca/1469804

This makes it look more like a bug with the way the accept() method is invoked inside PhptFilterIterator.

I tested with PHP 5.2 but do not see the same issue. I believe this is specific to PHP 5.3.

Below is a simplified testcase which I think exposes the same leak.
'/manyFiles' is a directory containing 10000 files.

<?php
$rii = new RecursiveIteratorIterator(
          new RecursiveDirectoryIterator('/manyFiles')
       );

function noop() {}

echo "noop call in loop (no leak)       : ";
foreach ($rii as $v) { noop($v); }
echo memory_get_usage() . PHP_EOL;

echo "strlen call in loop (leak on 53)  : ";
foreach ($rii as $v) { strlen($v); }
echo memory_get_usage() . PHP_EOL;
?>


Output on php52:
noop call in loop (no leak)       : 66176
strlen call in loop (leak on 53)  : 66176

Output on php53:
noop call in loop (no leak)       : 337448
strlen call in loop (leak on 53)  : 6028496

Looking backwards through PHP builds I have currently narrowed this down to a change that went in to PHP 5.3 somewhere between the 12th June 2008 and the 1st July 2008.

Testcase below shows that the issue relates to an implicit cast to string on an SplFileInfo object when retrieved from a RecursiveDirectoryIterator during iteration.

<?php
$rdi = new RecursiveDirectoryIterator('/manyFiles');

echo "SplFileInfo explicit cast to string: ";
foreach ($rdi as $v) { strlen((string)$v); } //OK
echo memory_get_usage(true) . PHP_EOL;

echo "SplFileInfo implicit cast to string: ";
foreach ($rdi as $v) { strlen($v); } //Leaky
echo memory_get_usage(true) . PHP_EOL;
?>

php52:
SplFileInfo explicit cast to string: 262144
SplFileInfo implicit cast to string: 262144

php53:
SplFileInfo explicit cast to string: 524288
SplFileInfo implicit cast to string: 6291456

A checkout of PHP53 from the 25th June 2008 does *not* have the problem

A checkout of PHP53 from the 26th June 2008 *does* have the problem

Here are the files that changed between those dates (I have taken out things that look irrelevant like test files)

RCS file: /repository/ZendEngine2/zend_builtin_functions.c,v
date: 2008/06/25 22:37:14;  author: felipe;  state: Exp;  lines: +2 -2
date: 2008/06/25 22:35:31;  author: felipe;  state: Exp;  lines: +1 -2

RCS file: /repository/php-src/ext/reflection/php_reflection.c,v
date: 2008/06/25 12:34:14;  author: dmitry;  state: Exp;  lines: +152 -2
date: 2008/06/25 12:33:46;  author: dmitry;  state: Exp;  lines: +128 -2

RCS file: /repository/php-src/ext/standard/formatted_print.c,v
date: 2008/06/25 10:16:52;  author: davidc;  state: Exp;  lines: +21 -25
date: 2008/06/25 08:56:42;  author: davidc;  state: Exp;  lines: +12 -4

RCS file: /repository/php-src/ext/standard/string.c,v
date: 2008/06/25 12:16:16;  author: ohill;  state: Exp;  lines: +251 -328

RCS file: /repository/php-src/main/main.c,v
date: 2008/06/25 12:18:51;  author: dmitry;  state: Exp;  lines: +3 -1
date: 2008/06/25 12:18:21;  author: dmitry;  state: Exp;  lines: +3 -1
date: 2008/06/25 12:18:36;  author: dmitry;  state: Exp;  lines: +3 -1

RCS file: /repository/php-src/main/php_ticks.c,v
date: 2008/06/25 12:18:51;  author: dmitry;  state: Exp;  lines: +6 -1
date: 2008/06/25 12:18:22;  author: dmitry;  state: Exp;  lines: +6 -1
date: 2008/06/25 12:18:36;  author: dmitry;  state: Exp;  lines: +6 -1

RCS file: /repository/php-src/main/php_ticks.h,v
date: 2008/06/25 12:18:51;  author: dmitry;  state: Exp;  lines: +2 -1
date: 2008/06/25 12:18:22;  author: dmitry;  state: Exp;  lines: +2 -1


The most likely candidates seem to be string.c and builtin_functions and it looks as though it is changes to parameter parsing that cause the problem. At this point I think we have done as much as we can to narrow this down. I'm changing this to Scripting engine as it clearly isn't an SPL issue.

Hi dmitry - please would you look at this?

The following patch fixes the issue. Might not be the nicest way and probably some other code in there might need a similar fix.

Index: Zend/zend_API.c
===================================================================
RCS file: /repository/ZendEngine2/zend_API.c,v
retrieving revision 1.296.2.27.2.34.2.64
diff -u -p -r1.296.2.27.2.34.2.64 zend_API.c
--- Zend/zend_API.c     4 Jun 2009 18:20:42 -0000       1.296.2.27.2.34.2.64
+++ Zend/zend_API.c     23 Jun 2009 21:33:04 -0000
@@ -254,10 +254,13 @@ ZEND_API int zend_get_object_classname(c
 static int parse_arg_object_to_string(zval **arg, char **p, int *pl, int type TSRMLS_DC) /* {{{ */
 {
        if (Z_OBJ_HANDLER_PP(arg, cast_object)) {
+               zval tmp;
+               INIT_PZVAL(&tmp);
                SEPARATE_ZVAL_IF_NOT_REF(arg);
-               if (Z_OBJ_HANDLER_PP(arg, cast_object)(*arg, *arg, type TSRMLS_CC) == SUCCESS) {
-                       *pl = Z_STRLEN_PP(arg);
-                       *p = Z_STRVAL_PP(arg);
+               if (Z_OBJ_HANDLER_PP(arg, cast_object)(*arg, &tmp, type TSRMLS_CC) == SUCCESS) {
+                       *pl = Z_STRLEN(tmp);
+                       *p = Z_STRVAL(tmp);
+                       zval_dtor(&tmp);
                        return SUCCESS;
                }
        }

This bug has been fixed in CVS.

Snapshots of the sources are packaged every three hours; this change
will be in the next snapshot. You can grab the snapshot at
http://snaps.php.net/.
 
Thank you for the report, and for helping us make PHP better.