Query on assert_options() api

Andy Wharmby wrote:

Hi Internals
A colleague of mine is currently working on creating new test
cases to improve the coverage of the PHP test cases and whilst
attempting to
write new tests for the assert extension hit on a problem when
attempting to query the current setting of ASSERT_CALLBACK. using the
assert_options() api.

With the following simple testcase :

<?php

function andy()
{
echo "andy called\n";
}

$o = assert_options(ASSERT_CALLBACK, "andy"); /* SET /
$n = assert_options(ASSERT_CALLBACK); / INQ */
echo "old setting is $o\n";
echo "new setting is $n\n"; ?>

The expected result was that the old (if any) and new setting of the
callback function name would be displayed but the actual results was

   old setting is 1
   new setting is 1

Looking at the code this is no surprise as the code in assert.c which
processes these requests unconditionally returns with RETURN_TRUE.
For all other assert options other than assert_options() function
returns the old value of the specified option as documented in the PHP
manual.

Further investigation uncovered that the code actually did behave in
the way we expected until it was changed to accept the array(&$obj,
"methodname")
syntax back in July 2001 (revision 1.32 of assert.c). At this time the
return was changed to be unconditionally TRUE.

Unfortunately this makes writing a test case to query the current
setting of the ASSERT_CALLBACK option impossible as assert_options()
no longer returns
any useable information for this option, i.e to code a testcase as
above that sets a value and then checks its set as expected.

If the code is modified as in the attached patch to instead return the
ZVAL for the ASSERT_CALLBACK option then we are able create the new
testcase to
verify the assert_options() api. However, I am concerned there is a
good reason this was not done when the code was changed back in
2001.Can anyone think
of a good reason why returning the zval on this api is not a good idea ?

Any comments on my proposed fix appreciated.

Regards
Andy

Andy Wharmby
IBM United Kingdom Limited

Hi All,
I have now completed further testing on the patch I posted
yesterday and have uncovered a further defect that will need addressing
if the
earlier fix is accepted.

With the code modified such that assert_options(ASSERT_CALLBACK) returns
the current setting of the option it uncovers a problem with
a recent change to assert.c to fix defect 39718 (
http://bugs.php.net/bug.php?id=39718).

With the following simple script

?php

function andy()
{
echo "andy called";
}

function default_callback()
{
echo "default_calback called";
}

/* Check php.ini setting set */
$o = assert_options(ASSERT_CALLBACK);
echo "Initial setting is "$o" ...it should be default_callback!!!\n";

assert(0);

/* modify callback */
$o= assert_options(ASSERT_CALLBACK, "andy");

assert(0);

?>

and the following php.ini setting:

assert.callback="default_callback"

then the expected result is:

Initial setting is "default_callback" ...it should be default_callback!!!
default_calback called

Warning: assert(): Assertion failed in
C:\Eclipse-PHP\workspace\Testcases\assertbug.php on line 16
andy called

Warning: assert(): Assertion failed in
C:\Eclipse-PHP\workspace\Testcases\assertbug.php on line 20

the actual result though is

Initial setting is "" ...it should be default_callback!!!
default_calback called

Warning: assert(): Assertion failed in
C:\Eclipse-PHP\workspace\Testcases\assert
bug.php on line 16
andy called

Warning: assert(): Assertion failed in
C:\Eclipse-PHP\workspace\Testcases\assert
bug.php on line 20

Rather than getting the php.ini setting of "default_callback" returned
NULL is returned instead. Issuing assert() calls the expected callback
OK though.

The problem is that the value of the global ASSERTG(cb) is not copied
to ASSERTG(callback) until the first call to assert() by a request
so if the script includes a query on the setting BEFORE the first call
to assert() then the value returned is the default INI setting, i.e.
NULL, rather
than any value specified in php.ini. The issues is easily addressed by
moving the code that populates ASSERTG(callback) to a new RINIT function
in assert.c.

The modified patches for all the necessary changes to assert.c and
basic_functions.c are attached. If the fix is accepted I will get my
colleague to drop the
updated PHPT tests into CVS.

If a defect needs to be opened for this issue just let me know and I
will do so.

Regards
Andy

Andy Wharmby
IBM United Kingdom Limited