When PDOStatement->execute() failed, then execute successfully, errorInfo() information is incorrect

[Yurunsoft]

Description

The following code:

<?php
$pdo = new PDO('mysql:host=127.0.0.1;dbname=db_imi_test', 'root', 'root', [
    PDO::ATTR_ERRMODE => PDO::ERRMODE_SILENT,
]);
$pdo->setAttribute(PDO::ATTR_ERRMODE, PDO::ERRMODE_SILENT);
$stmt = $pdo->prepare('insert into tb_test(value) values(?)');
// fail
var_dump($stmt->execute([\PHP_INT_MIN]), $stmt->errorCode(), $stmt->errorInfo());

// success
var_dump($stmt->execute([1]), $stmt->errorCode(), $stmt->errorInfo());

Resulted in this output:

bool(false)
string(5) "22003"
array(3) {
  [0]=>
  string(5) "22003"
  [1]=>
  int(1264)
  [2]=>
  string(46) "Out of range value for column 'value' at row 1"
}
bool(true)
string(5) "00000"
array(3) {
  [0]=>
  string(5) "00000"
  [1]=>
  int(1264)
  [2]=>
  string(46) "Out of range value for column 'value' at row 1"
}

But I expected this output instead:

bool(false)
string(5) "22003"
array(3) {
  [0]=>
  string(5) "22003"
  [1]=>
  int(1264)
  [2]=>
  string(46) "Out of range value for column 'value' at row 1"
}
bool(true)
string(5) "00000"
array(3) {
  [0]=>
  string(5) "00000"
  [1]=>
  NULL
  [2]=>
  NULL
}

PHP Version

PHP 7.4、8.0、8.1

Operating System

Ubuntu 20.04

[twose]

I don't think it's a bug, errorInfo is more like lastErrorInfo (or errno) here, you should check the return value of the operations before you access the errorInfo.
But the PR looks OK for me, it skips the fetch_err() call when there is no error, just like 307c1f6 did.

[cmb69]

This looks clearly like a bug to me, because the result of ::errorInfo() is inconsistent. We need to be careful, though, not to break some drivers when fixing this.

[twose]

This looks clearly like a bug to me, because the result of ::errorInfo() is inconsistent. We need to be careful, though, not to break some drivers when fixing this.

Inconsistent behavior is a problem indeed, that's why I think this patch does make sense in php-src. In this sense, it is indeed a BUG.

But I'm sorry that I have not pointed out a context that is not in here:
There is a script that did not check the return value of execute() but checked errorInfo instead, then, this problem was discovered. I definitely do not advocate such an approach in any case.
The question is also linked to Swoole here. In Swoole, It is more like such a question: do we need to clear the last error after DB queries every time.
IMO, this change will not affect the behavior of any good practice, except for tests.