Original patch notes from #45259

I have experienced this bug also. I tried most of the windows builds which are available on the download page: 5.3.4 (ts and nts, vc6 and vc9) and 5.2.16 (ts and nts).

My tests try to execute a prepared statement with parameters and what I'd like PDO to do is map my php variable to a PDO_PARAM_* type. This is where things go wrong.

Example:
$db = new PDO('sqlite::memory:');

$stmt = $db->prepare('SELECT "string" = ? AS test');
$stmt->execute(array('string'));
$row = $stmt->fetch();
echo 'test: ', $row['test'], PHP_EOL;
// prints "test: 1", I expect "test: 1"

$stmt = $db->prepare('SELECT 50 = ? AS test');
$stmt->execute(array(50));
$row = $stmt->fetch();
echo 'test: ', $row['test'], PHP_EOL;
// prints "test: 0", I expect "test: 1"

The strange thing is:
$stmt = $db->prepare('SELECT 50 = 50 AS test');
$stmt->execute();
$row = $stmt->fetch();
echo 'test: ', $row['test'], PHP_EOL;
// prints "test: 1", I expect "test: 1"

This indicates sqlite knows what I mean, but somewhere in PDO things go bump.

I've investigated by looking into the source (SVN 5.3 rev 306430 from yesterday on a fresh Debian lenny install). And devised a few test cases and a patch. In the pdo_sqlite extension it all works peachy, although for the mapping of ZVAL to PDO_PARAM_* I've had to edit the pdo_stmt.c file in the pdo extension. I don't know if this is acceptable as I can imagine many drivers depend on the pdo extension being frozen.

I've added an exception to the PDOStatement::execute() handler, the handler sets the default PDO_PARAM_* to PDO_PARAM_ZVAL to signal that no explicit type was set. So the driver can do it's own mapping. This only takes effect when the driver for the statment is sqlite. Others might depend on the default to be PDO_PARAM_STR.
This cannot stay PDO_PARAM_STR because really_register_bound_param() will try and autoconvert the ZVAL if it is PDO_PARAM_STR or a couple of other param types. I assume this will destroy the original ZVAL's type and will make it indistinguishable from an explicitly defined PDO_PARAM_STR param.

So now the pdo_sqlite_stmt_param_hook() will receive a parameter with PDO type PDO_PARAM_ZVAL when it is not explicitly set by the user, this will allow the driver to inspect the given zval and bind it accordingly. This has also been included in the patch which are alterations of only the pdo_sqlite_stmt_param_hook() function. I've revised it a bit, it looked odd to me all those return statements scattered throughout the nested switch statements.

I've only tested this with the SVN build from yesterday on Debian lenny x86. Maybe we should test it on more platforms? And can someone comment on the patch, especially if we can merge it into the svn 5.3 branch?

Behavior confirmed: <https://3v4l.org/LsC9s>.

However, I don't think this is a bug (since it's possible to call
::bindValue() explicitly), but rather a missing feature.  Given
that a special casing in PDOStatement is suggested, discussion on
the internals@ mailing appears to be appropriate.