If you manually change HAVE_BROKEN_GETCWD in php_config.h to 
#define HAVE_BROKEN_GETCWD 1
after configure and do 'make clean && make', does it really work then..?

Hi again!

(Thanks for fixing the mpm_common crash, that problem is gone now.)

All the following with #define HAVE_BROKEN_GETCWD 1 in main/php_config.h, and the following code (tests 1-4):

<?php

  print getcwd() . "<BR>";

  $handle = fopen("a.txt", "w", false);
  if ( $handle != FALSE ) {
    fputs($handle, "testtext");
    fclose($handle);
  }

?>

..and, respectively (tests 5-8):

<?php

  print getcwd() . "<BR>";

  $handle = fopen("./a.txt", "w", false);
  if ( $handle != FALSE ) {
    fputs($handle, "testtext");
    fclose($handle);
  }

?>


1. No existing "a.txt" file in the destination directory. With the "read" flag missing to the "test" directory along the path, and write permissions to the destination directory:

Warning: fopen(): open_basedir restriction in effect. File(a.txt) is not within the allowed path(s): (.:/apache/php/lib/php/:/apache/htdocs/bwiberg/) in /apache/htdocs/bwiberg/test/safemode/write.php on line 5 Warning: fopen(a.txt): failed to open stream: Not owner in /apache/htdocs/bwiberg/test/safemode/write.php on line 5

No file gets created. getcwd() fails.


2. No existing "a.txt" file in the destination directory. Having both read and execute flags along the path, and write permissions to the destination directory:

/apache/htdocs/bwiberg/test/safemode
Warning: fopen(): Unable to access a.txt in /apache/htdocs/bwiberg/test/safemode/write.php on line 5 Warning: fopen(a.txt): failed to open stream: No such file or directory in /apache/htdocs/bwiberg/test/safemode/write.php on line 5 

No file gets created. But getcwd() correctly returns "
/apache/htdocs/bwiberg/test/safemode".


3. No existing "a.txt" file in the destination directory. Trying with "./a.txt" instead, a missing "read" flag to the "test" directory along the path, and write permissions to the destination directory:

No error message. The a.txt file gets correctly created, but getcwd() fails.


4. No existing "a.txt" file in the destination directory. Trying with "./a.txt", both "read" and "execute" flags along the path, and write permissions to the destination directory:

No error message. The file gets correctly created, and getcwd() returns "/apache/htdocs/bwiberg/test/safemode".


5. Existing "a.txt" file in the destination directory. With the "read" flag missing to the "test" directory along the path, and write permissions to the destination directory:

Warning: fopen(): open_basedir restriction in effect. File(a.txt) is not within the allowed path(s): (.:/apache/php/lib/php/:/apache/htdocs/bwiberg/) in /apache/htdocs/bwiberg/test/safemode/write.php on line 5 Warning: fopen(a.txt): failed to open stream: Not owner in /apache/htdocs/bwiberg/test/safemode/write.php on line 5

File does not get overwritten. getcwd() fails.


6. Existing "a.txt" file in the destination directory. Having both read and execute flags along the path, and write permissions to the destination directory:

No error message. File gets overwritten correctly. getcwd() returns "/apache/htdocs/bwiberg/test/safemode".


7. Existing "a.txt" file in the destination directory. Trying with "./a.txt" instead, a missing "read" flag to the "test" directory along the path, and write permissions to the destination directory:

No error message. File gets overwritten correctly, but getcwd() fails.

8. Existing "a.txt" file in the destination directory. Trying with "./a.txt", both "read" and "execute" flags along the path, and write permissions to the destination directory:

No error message. File gets overwritten correctly. getcwd() returns "/apache/htdocs/bwiberg/test/safemode".


From a getcwd() perspective (the scope of this particular bug report), tests number 1, 2, 3, 5 and 7 fail although they perhaps shouldn't.

So, it appears that HAVE_BROKEN_GETCWD does not solve the "problems" on AIX (with the current code for HAVE_BROKEN_GETCWD). Hence, the proposed change to acinclude.m4 is currently not needed.

Any ideas on how to go from here? Or should we accept that getcwd() should fail under these circumstances?

Thanks in advance!

Best regards,
Bj?rn

We just accept it. As long as we don't have IBM sponsoring us AIX machines to test with I won't fix these issues.