Changeset 18995 can lead to Notice: Undefined index: wp_the_query

[bobbingwide]

In changeset r18995 there were two changes to query.php
The comment "new does not require by reference" applies to the first but not the second change.

Should the assignment by reference in wp_reset_query have been removed?

The reported problem is:
​http://wordpress.org/support/topic/plugin-wordpress-seo-by-yoast-notice-undefined-index-wp_the_query?replies=2#post-2477593

I had the same problem and questioned the suggested fix.
What I don't understand is why there is no Notice: message for the original code.

Try running this code.

<?php
error_reporting( E_ALL );  
@ini_set('display_errors',1);

function bad() {
  print_r( __FUNCTION__ );
  echo "\n";

  $GLOBALS = array( 'wp_query' => 'set' );
  print_r( $GLOBALS );
  unset($GLOBALS['wp_query']);
  $GLOBALS['wp_query'] = $GLOBALS['wp_the_query'];
  print_r( $GLOBALS );
  $GLOBALS['wp_query'] =& $GLOBALS['wp_the_query'];
  print_r( $GLOBALS );
}

function good() {
  print_r( __FUNCTION__ );
  echo "\n";
  $GLOBALS = array( 'wp_query' => 'set' );
  print_r( $GLOBALS );
  unset($GLOBALS['wp_query']);
  $GLOBALS['wp_query'] =& $GLOBALS['wp_the_query'];
  print_r( $GLOBALS );
  $GLOBALS['wp_query'] = $GLOBALS['wp_the_query'];
  print_r( $GLOBALS );
}

good();
bad();

[SergeyBiryukov]

19412.patch​ adds a _doing_it_wrong() message if wp_reset_query() was called before query_posts().

[nacin]

Per a discussion in IRC earlier today, I think it would be better to just close this as wontfix.

• If $wp_the_query is an object, then everything will work.

• If $wp_the_query is not yet set, then nothing will break (correct me if I'm wrong) and a notice will result.

At this point, if you're using it wrong, you get a notice. Seems fine to me. The alternatives are A) return silently if $wp_the_query isn't set, or B) issue a _doing_it_wrong() notice. B) is kind of redundant since a notice is already being returned, and A) hides from developers that there is a problem.

Has this been seen outside of Yoast's plugin? I imagine he'll be quick to fix it.

[ryan]

Agreed.

[bobbingwide]

I didn't see the IRC discussion but I would have opted for alternative A.
IMHO it's better defensive coding.
Who says there's a problem if wp_reset_query is called before query_posts().

And it still doesn't explain why PHP masked the Notice when the & was present.

[bobbingwide]

May I refer you to this blog post

​http://nacin.com/2010/04/23/5-ways-to-debug-wordpress/

it contains
WP_DEBUG will often reveal potential problems in your code, such as unchecked indexes (empty() and isset() are your friend) and undefined variables. (You may even find problems in WordPress itself, in which case you should file a bug report.)

[SergeyBiryukov]

Replying to bobbingwide:

I didn't see the IRC discussion

​https://irclogs.wordpress.org/chanlog.php?channel=wordpress-dev&day=2011-12-02&sort=asc#m341113

[bobbingwide]

Thanks I just read the discussion as I was writing something about the shortcodes and wpautop problem #10457

PS. Shouldn't there be a similar change in wp-settings.php?
$wp_query =& $wp_the_query;

[nacin]

Replying to bobbingwide:

And it still doesn't explain why PHP masked the Notice when the & was present.

Because it creates a reference between variables, rather than any sort of assignment. So you can say $a =& $b; without $b being set, because then $b references $a (or vice versa). However, objects are automatically passed by reference, which means that if $b is an object, then $a = $b; makes them both reference the same object.

Where the notice kicks in is where $wp_the_query isn't yet defined. There is no requirement for query_posts() to have been called — the issue is specifically calling a query-related function before the query is even initialized in wp-settings.php.

(You may even find problems in WordPress itself, in which case you should file a bug report.)

Thank you for referring me to one of my own blog posts. :-)

The change in wp-settings.php could also occur.

[scribu]

Related: #22125

[wonderboymusic]

For that lingering ref, fixed here: [22434]