Fix error message allocation of PDO PgSQL.

[ranvis]

In PDO PgSQL driver, some error messages are allocated using estrdup() instead of pestrdup().
This cause crash if persistent connection is used and the last error message is freed (with pefree().)

<?php
$db = new PDO(
    'pgsql:dbname=' . DB_NAME,
    USER_NAME,
    PASSWORD,
    [
        PDO::ATTR_PERSISTENT => true,
        PDO::ATTR_ERRMODE => PDO::ERRMODE_EXCEPTION,
        PDO::ATTR_EMULATE_PREPARES => false,
    ],
);

$st = $db->prepare('select 1');
for ($i = 0; ++$i <= 2;) {
    try {
        $st->bindValue(':invalid', $i);
    } catch (PDOException $e) {
        echo "$i\n";
    }
}

Expected result:

1
2

Actual result:

1
*** Error in `php': free(): invalid pointer: 0x... ***
======= Backtrace: =========
/lib64/libc.so.6(+0x81329)[0x...]
/usr/lib64/php/modules/pdo_pgsql.so(+0x3c11)[0x...]

[cmb69]

Thank you for the PR! The fix is correct, but it would be good to have a regression tests for that. You can copy the boilerplate from any of the pgo_pgsql tests, for instance, https://github.com/php/php-src/blob/PHP-8.0/ext/pdo_pgsql/tests/bug73959.phpt.

[ranvis]

Added regression test as pr7723.phpt. Thank you for the pointer.

Regarding #7725, I first wondered if Issue must be created first.
Another thing that I wondered is that I cannot find a naming convention for *.phpt compatible with the current system.
https://qa.php.net/write-test.php#naming-conventions
I know bug{id}.phpt is bad because the number will collide in future though.

[cmb69]

Thanks for adding the regression test! And, frankly, I haven't thought about the naming issue so far. Maybe we should just go with gh7723.phpt; maybe that should be discussed on the mailing list. Maybe @nikic has some thoughts about that.

[ranvis]

Thanks. Not sure how it will go I changed to the better one anyway.