Registration is now based solely on object type, which allows the code around this to be simplified significantly.
In the process of making this adjustment:
* `register_meta()`, `unregister_meta_key()`, `get_registered_metadata()`, and `registered_meta_key_exists()` no longer return `WP_Error` objects.
* The recently introduced `wp_object_type_exists()` function and the restriction on object type has been removed.
Note: No guarantee of uniqueness is made across object subtypes. Registered meta keys should be uniquely prefixed to avoid conflict.
Fixes#35658.
Built from https://develop.svn.wordpress.org/trunk@38095
git-svn-id: http://core.svn.wordpress.org/trunk@38036 1a063a9b-81f0-0310-95a4-ce76da25c4cd
When using `register_meta()` with the function signature from 4.5 and earlier, the `auth_{$type}_meta_{$key}` and `sanitize_{$type}_meta_{$key}` filters are used. Any calls to `register_meta()` expecting this behavior should continue to work. The new filters, which take advantage of object subtypes, should not be added unless the proper `$args` array is passed.
See #35658.
Built from https://develop.svn.wordpress.org/trunk@38041
git-svn-id: http://core.svn.wordpress.org/trunk@37982 1a063a9b-81f0-0310-95a4-ce76da25c4cd
If auth and/or sanitize callbacks are specified in the arguments for
`register_meta()`, filters are added to handle these callbacks. These
should be removed when calling `unregister_meta_key()` to avoid
unintentional filtering.
See #35658.
Built from https://develop.svn.wordpress.org/trunk@38040
git-svn-id: http://core.svn.wordpress.org/trunk@37981 1a063a9b-81f0-0310-95a4-ce76da25c4cd
In doing this, non-core object types are no longer forcibly blocked and are instead checked against `wp_object_type_exists()` which has a filterable return value. Still, filter that at your own risk.
props Faison for the initial patch.
see 35658.
Built from https://develop.svn.wordpress.org/trunk@37991
git-svn-id: http://core.svn.wordpress.org/trunk@37932 1a063a9b-81f0-0310-95a4-ce76da25c4cd
`register_meta()` has been altered to accept an array of arguments as the third parameter in order to support its usage beyond XML-RPC, notably in the REST API and other projects that may build on top of meta, such as a potential Fields API. Arguments are whitelisted to reserve the right for core to add more later.
New functions added to complement this expansion are:
* `registered_meta_key_exists()`
* `unregister_meta_key()`
* `get_registered_meta_keys()`
* `get_registered_metadata()`
* A "private" function for the aforementioned whitelisting.
There still need to be lots of tests written for previous and new behaviors, and many things are subject to change. Maybe things will explode. #yolo
props jeremyfelt, ericlewis, sc0ttkclark, helen, rmccue, ocean90, voldemortensen.
see #35658.
Built from https://develop.svn.wordpress.org/trunk@37924
git-svn-id: http://core.svn.wordpress.org/trunk@37865 1a063a9b-81f0-0310-95a4-ce76da25c4cd
Comment and term meta lazyloading for `WP_Query` loops, introduced in 4.4,
depended on filter callback methods belonging to `WP_Query` objects. This meant
storing `WP_Query` objects in the `$wp_filter` global (via `add_filter()`),
requiring that PHP retain the objects in memory, even when the local variables
would typically be expunged during normal garbage collection. In cases where a
large number of `WP_Query` objects were instantiated on a single pageload,
and/or where the contents of the `WP_Query` objects were quite large, serious
performance issues could result.
We skirt this problem by moving metadata lazyloading out of `WP_Query`. The
new `WP_Metadata_Lazyloader` class acts as a lazyload queue. Query instances
register items whose metadata should be lazyloaded - such as post terms, or
comments - and a `WP_Metadata_Lazyloader` method will intercept comment and
term meta requests to perform the cache priming. Since `WP_Metadata_Lazyloader`
instances are far smaller than `WP_Query` (containing only object IDs), and
clean up after themselves far better than the previous `WP_Query` methods (bp
only running their callbacks a single time for a given set of queued objects),
the resource use is decreased dramatically.
See [36525] for an earlier step in this direction.
Props lpawlik, stevegrunwell, boonebgorges.
Fixes#35816.
Built from https://develop.svn.wordpress.org/trunk@36566
git-svn-id: http://core.svn.wordpress.org/trunk@36533 1a063a9b-81f0-0310-95a4-ce76da25c4cd
The `$delete_all` flag in `delete_metadata()` triggers cache invalidation for
multiple objects. Previously, invalidation took place for all objects matching
the `$meta_key` parameter, regardless of whether `$meta_value` was also set.
This resulted in overly aggressive invalidation.
Props rahal.aboulfeth.
Fixes#35797.
Built from https://develop.svn.wordpress.org/trunk@36511
git-svn-id: http://core.svn.wordpress.org/trunk@36478 1a063a9b-81f0-0310-95a4-ce76da25c4cd
A meta_query containing an `OR` relation can result in the same record matching
multiple clauses, leading to duplicate results. The previous prevention against
duplicates [18178] #17582 became unreliable in 4.1 when `WP_Meta_Query`
introduced support for nested clauses. The current changeset adds a new method
`WP_Meta_Query::has_or_relation()` for checking whether an `OR` relation
appears anywhere in the query, and uses the new method in `WP_User_Query` to
enforce distinct results as necessary.
Props maxxsnake.
Fixes#32592.
Built from https://develop.svn.wordpress.org/trunk@32713
git-svn-id: http://core.svn.wordpress.org/trunk@32683 1a063a9b-81f0-0310-95a4-ce76da25c4cd
The implementation of `meta_query` orderby introduced in [31312] put clause
identifiers into a 'name' parameter of the clause. For greater clarity, this
changeset updates the syntax to use the associative array key used when
defining `meta_query` parameters, instead of the 'name' parameter.
Props Funkatronic, DrewAPicture.
Fixes#31045.
Built from https://develop.svn.wordpress.org/trunk@31340
git-svn-id: http://core.svn.wordpress.org/trunk@31321 1a063a9b-81f0-0310-95a4-ce76da25c4cd
`WP_Meta_Query` clauses now support a 'name' parameter. When building a
`WP_Query` object, the value of 'orderby' can reference this 'name', so that
it's possible to order by any clause in a meta_query, not just the first one
(as when using 'orderby=meta_value'). This improvement also makes it possible
to order by multiple meta query clauses (or by any other eligible field plus
a meta query clause), using the array syntax for 'orderby' introduced in [29027].
Props Funkatronic, boonebgorges.
Fixes#31045.
Built from https://develop.svn.wordpress.org/trunk@31312
git-svn-id: http://core.svn.wordpress.org/trunk@31293 1a063a9b-81f0-0310-95a4-ce76da25c4cd
It's necessary for hook docs to directly precede hook lines so the parser can correctly match them up. This change relocates hook docs for `update_{$meta_type}_meta`, `update_postmeta`, `updated_{$meta_type}_meta`, and `updated_postmeta`.
Fixes#11683.
Built from https://develop.svn.wordpress.org/trunk@30149
git-svn-id: http://core.svn.wordpress.org/trunk@30149 1a063a9b-81f0-0310-95a4-ce76da25c4cd
Per our inline documentation standards, no further use of the `@uses` tag is recommended as used and used-by relationships can be derived through other means. This removes most uses of the tag in core documentation, with remaining tags to be converted to `@global` or `@see` as they apply.
Fixes#30191.
Built from https://develop.svn.wordpress.org/trunk@30105
git-svn-id: http://core.svn.wordpress.org/trunk@30105 1a063a9b-81f0-0310-95a4-ce76da25c4cd
Passing orderby=meta_value and meta_key=foo to WP_Query should require that
each matched post have a meta value with the key 'foo'. To make this
requirement compatible with meta_query params that have the relation OR, we
nest the meta_query param, and join it using AND to a meta_query clause
generated from the meta_key/meta_compare/meta_type query vars.
Fixes#25538.
Built from https://develop.svn.wordpress.org/trunk@29964
git-svn-id: http://core.svn.wordpress.org/trunk@29711 1a063a9b-81f0-0310-95a4-ce76da25c4cd
The logic used to generate clause SQL in `WP_Meta_Query` is somewhat arcane,
stemming mostly from an ongoing effort to eliminate costly table joins when
they are not necessary. By systematizing the process of looking for shareable
joins - as was done in `WP_Tax_Query` [29902] - it becomes possible to simplify
the construction of SQL queries in `get_sql_for_clause()`. Moreover, the
simplified logic is actually considerably better at identifying shareable
joins, such that certain uses of `WP_Meta_Query` will see joins reduced by 50%
or more.
Includes integration tests for a representative cross-section of the query
clause combinations that result in shared table aliases.
Props boonebgorges, sc0ttkclark.
See #24093.
Built from https://develop.svn.wordpress.org/trunk@29940
git-svn-id: http://core.svn.wordpress.org/trunk@29691 1a063a9b-81f0-0310-95a4-ce76da25c4cd
Previously, meta query arguments could be joined by a single AND or OR relation.
Now, these queries can be arbitrarily nested, allowing clauses to be linked
together with multiple relations.
Adds unit tests for the new nesting syntax. Modifies a few existing unit tests
that were overly specific for the old SQL syntax. Backward compatibility with
existing syntax is fully maintained.
Props boonebgorges, DrewAPicture.
See #29642.
Built from https://develop.svn.wordpress.org/trunk@29887
git-svn-id: http://core.svn.wordpress.org/trunk@29643 1a063a9b-81f0-0310-95a4-ce76da25c4cd
The exceptions to this are update_post_meta() and add_post_meta() which are often used by plugins in POST handlers and will continue accepting slashed data for now.
Introduce wp_upate_post_meta() and wp_add_post_meta() as unslashed alternatives to update_post_meta() and add_post_meta(). These functions could become methods in WP_Post so don't use them too heavily yet.
Remove all escape() calls from wp_xmlrpc_server. Now that core expects unslashed data this is no longer needed.
Remove addslashes(), addslashes_gpc(), add_magic_quotes() calls on data being prepared for handoff to core functions that until now expected slashed data. Adding slashes in no longer necessary.
Introduce wp_unslash() and use to it remove slashes from GPCS data before using it in core API. Almost every instance of stripslashes() in core should now be wp_unslash(). In the future (a release or three) when GPCS is no longer slashed, wp_unslash() will stop stripping slashes and simply return what is passed. At this point wp_unslash() calls can be removed from core.
Introduce wp_slash() for slashing GPCS data. This will also turn into a noop once GPCS is no longer slashed. wp_slash() should almost never be used. It is mainly of use in unit tests.
Plugins should use wp_unslash() on data being passed to core API.
Plugins should no longer slash data being passed to core. So when you get_post() and then wp_insert_post() the post data from get_post() no longer needs addslashes(). Most plugins were not bothering with this. They will magically start doing the right thing. Unfortunately, those few souls who did it properly will now have to avoid calling addslashes() for 3.6 and newer.
Use wp_kses_post() and wp_kses_data(), which expect unslashed data, instead of wp_filter_post_kses() and wp_filter_kses(), which expect slashed data. Filters are no longer passed slashed data.
Remove many no longer necessary calls to $wpdb->escape() and esc_sql().
In wp_get_referer() and wp_get_original_referer(), return unslashed data.
Remove old stripslashes() calls from WP_Widget::update() handlers. These haven't been necessary since WP_Widget.
Switch several queries over to prepare().
Expect something to break.
Props alexkingorg
see #21767
git-svn-id: http://core.svn.wordpress.org/trunk@23416 1a063a9b-81f0-0310-95a4-ce76da25c4cd
You could already use EXISTS by omitting a value to check.
props georgestephanis, scribu
fixes#18158
git-svn-id: http://core.svn.wordpress.org/trunk@21185 1a063a9b-81f0-0310-95a4-ce76da25c4cd
