A number of the HTTP external tests can inconstantly fail. As the HTTP API is one that doesn't change often, this failure creates noise. With the goal of increasing the signal from the unit tests, these tests are now skipped if they timeout. A notice is added when running the external http tests so that the developer knows what skipped tests may mean here.
See #33968
git-svn-id: https://develop.svn.wordpress.org/trunk@34848 602fd350-edb4-49c9-b593-d223f7449a82
Adds unit tests for `WP_Customize_Setting::save()` (and `WP_Customize_Setting::update()`), along with the actions `customize_update_{$type}`, and `customize_save_{$id_base}` which they trigger.
Fixes#34140.
git-svn-id: https://develop.svn.wordpress.org/trunk@34838 602fd350-edb4-49c9-b593-d223f7449a82
The `WP_INSTALLING` constant is a flag that WordPress sets in a number of
places, telling the system that options should be fetched directly from the
database instead of from the cache, that WP should not ping wordpress.org for
updates, that the normal "not installed" checks should be bypassed, and so on.
A constant is generally necessary for this purpose, because the flag is
typically set before the WP bootstrap, meaning that WP functions are not yet
available. However, it is possible - notably, during `wpmu_create_blog()` -
for the "installing" flag to be set after WP has already loaded. In these
cases, `WP_INSTALLING` would be set for the remainder of the process, since
there's no way to change a constant once it's defined. This, in turn, polluted
later function calls that ought to have been outside the scope of site
creation, particularly the non-caching of option data. The problem was
particularly evident in the case of the automated tests, where `WP_INSTALLING`
was set the first time a site was created, and remained set for the rest of the
suite.
The new `wp_installing()` function allows developers to fetch the current
installation status (when called without any arguments) or to set the
installation status (when called with a boolean `true` or `false`). Use of
the `WP_INSTALLING` constant is still supported; `wp_installing()` will default
to `true` if the constant is defined during the bootstrap.
Props boonebgorges, jeremyfelt.
See #31130.
git-svn-id: https://develop.svn.wordpress.org/trunk@34828 602fd350-edb4-49c9-b593-d223f7449a82
This function attempts to read from the relationship cache, and uses any data
it finds. If it finds no data, it does a query for the data it needs. Since we
are going to the trouble to query for the relationships, and since we are
already using cached data when available, let's go ahead and cache it for
later use.
Props joehoyle, boonebgorges.
Fixes#32044.
git-svn-id: https://develop.svn.wordpress.org/trunk@34812 602fd350-edb4-49c9-b593-d223f7449a82
Renames `reset_permalinks()` to `set_permalink_structure()` (mimicking `$wp_rewrite->set_permalink_structure()`) and allows it to accept an optional permalink structure. In this way, we can double dip using it to both set and reset the permalink structure from anywhere.
Removes alot of duplicated code from tests.
See #33968.
git-svn-id: https://develop.svn.wordpress.org/trunk@34810 602fd350-edb4-49c9-b593-d223f7449a82
`wp_insert_term()` doesn't allow the creation of a term when the term `name`
is the same as another term in the same hierarchy level of the same taxonomy.
Previously, this duplicate check used `get_term_by( 'name' )`, which uses the
database collation to determine sameness. But common collations do not
distinguish between accented and non-accented versions of a character. As a
result, it was impossible to create a term 'Foo' if a sibling term with an
accented character existed.
We address this problem by using `get_terms()` to do the duplicate check. This
query returns all potentially matching terms. We then do a stricter check
for equivalence in PHP, before determining whether one of the matches is
indeed a duplicate.
Props boonebgorges, tyxla, geza.miklo, mehulkaklotar.
Fixes#33864.
git-svn-id: https://develop.svn.wordpress.org/trunk@34809 602fd350-edb4-49c9-b593-d223f7449a82
Removes the restoration logic, which leveraged a static property initialized with `get_option()`.
See #33968.
git-svn-id: https://develop.svn.wordpress.org/trunk@34807 602fd350-edb4-49c9-b593-d223f7449a82
The function now uses the following order of precedence when calculating
comment pagination: 1. the 'per_page' value passed in the `$args` array,
2. the 'comments_per_page' query var in `$wp_query`, and 3. the
'comments_per_page' setting from options-discussion.php. This change allows
`get_page_of_comment()` to return an accurate value before the main query
has been run.
Props laceous.
See #13939.
git-svn-id: https://develop.svn.wordpress.org/trunk@34806 602fd350-edb4-49c9-b593-d223f7449a82
This prevents notices from being thrown when a 'pre_get_users' callback
removes required values from the list of `query_vars`.
For backward compatibility with previous uses of 'pre_get_users', default
values are parsed both before and after the action is fired.
Fixes#33449.
git-svn-id: https://develop.svn.wordpress.org/trunk@34804 602fd350-edb4-49c9-b593-d223f7449a82
Introduces `add_network_option()`, `update_network_option()`, `get_network_option()`, and `delete_network_option()`, which use the internals previously found in the `*_site_option()` functions and allow for a network ID to enable cross network storage and retrieval of network options.
* `*_site_option()` functions are now wrappers for `*_network_option()` and will interact with options for the current network in a multisite environment.
* All will continue to fallback to `*_option()` for non-multisite uses.
* Adds basic tests for cross network storage and retrieval.
* Existing tests for `*_site_option()` functions provide coverage for storage and retrieval of current network options.
Props spacedmonkey, DrewAPicture, jeremyfelt.
See #28290.
git-svn-id: https://develop.svn.wordpress.org/trunk@34777 602fd350-edb4-49c9-b593-d223f7449a82
After [33807], `Tests_Image_Intermediate_Size::test_get_intermediate_sizes_by_array_zero_width()`
was failing intermittently. This failure was due to the use of a random number
for the image height. When the height was sufficiently large - $height >= 202 -
to change the aspect ratio of the cropped image (based on WP's threshold of a
1px difference), the test passed. And when the height was exactly 200, the
medium image size was hit exactly. The failure occurred only with a height of
201, which is close enough to 200 so that WP determined that the aspect ratio
of the potential crop was close enough to match.
The current changeset splits the test into two, in order to demonstrate the
failure.
See #17626. Fixes#34087.
git-svn-id: https://develop.svn.wordpress.org/trunk@34775 602fd350-edb4-49c9-b593-d223f7449a82
* Add test file accidentally omitted from [34761].
* Bail properly from benchmarker when a `preg_last_error()` is found.
Props miqrogroove.
See #34121.
git-svn-id: https://develop.svn.wordpress.org/trunk@34773 602fd350-edb4-49c9-b593-d223f7449a82
In versions of PHP earlier than 5.4, the `$matches` parameter was required.
Excluding it here meant that PCRE benchmarking tests were erroring on older
versions of PHP without cleaning up the PCRE ini settings, causing various
problems on later tests.
Props miqrogroove, dd32, boonebgorges.
And to the cosmic forces that cause bugs like this. The best part of waking up
is diagnosing leakage between automated tests.
See #34121.
git-svn-id: https://develop.svn.wordpress.org/trunk@34772 602fd350-edb4-49c9-b593-d223f7449a82
* Move pattern from `wptexturize()` into a separate function.
* Move pattern from `wp_html_split()` into a separate function.
* Beautify code for `wp_html_split()`.
* Remove unnecessary instances of `/s` modifier in patterns that don't use dots.
* Add `tests/phpunit/data/formatting/whole-posts.php` for testing larger strings.
* Add function `benchmark_pcre_backtracking()`.
* Add tests for `wp_html_split()`.
* Add tests for `wptexturize()`.
* Add tests for `get_shortcode_regex()`.
Props miqrogroove.
Fixes#34121.
git-svn-id: https://develop.svn.wordpress.org/trunk@34761 602fd350-edb4-49c9-b593-d223f7449a82
Specifically, `::insert()`, `::replace()`, `::update()`, and `::delete()` can now set a column to `NULL`, or add the `IS NULL` condition to the `WHERE` clause.
This is based on [backpress 279].
Props pento, nbachiyski, sorich87.
Fixes#15158.
git-svn-id: https://develop.svn.wordpress.org/trunk@34737 602fd350-edb4-49c9-b593-d223f7449a82
After [34561], `wp_list_comments()` no longer passed all of a post's comments
to `Walker_Comments`. As a result, calls to `get_comment_link()` occurring
inside the comment loop had insufficient context to determine the proper
'cpage' value to use when generating comment permalinks. This, in turn, caused
comment permalinks to behave erratically.
The current changeset addresses the problem as follows:
* `get_comment_link()` now accepts a 'cpage' parameter. When present, 'cpage' will be used to build the comment permalink - no automatic calculation will take place.
* When called within the main loop, `wp_list_comments()` calculates the proper 'cpage' value for comments in the loop, and passes it down to `get_comment_link()`.
* `cpage` and `comment-page-x` query vars are generally required in comment permalinks (see #34068), but an exception is made when 'default_comment_page=oldest': the bare post permalink will always be the same as `cpage=1`, so `cpage` is excluded in this case.
Props peterwilsoncc for assiduous spreadsheeting.
Fixes#34073.
git-svn-id: https://develop.svn.wordpress.org/trunk@34735 602fd350-edb4-49c9-b593-d223f7449a82
[34561] "fixed" the problem of newest-first comments showing fewer than
'per_page' comments on the post permalink when the total number of comments
was not divisible by 'per_page'. See #29462. But this fix caused numerous
other problems. First, comment pages reported by `get_page_of_comment()`
(which expects comment pages to be filled oldest-first) were no longer correct.
Second, and more seriously, the new logic caused comments to be shifted
between pages, making their permalinks non-permanent.
The current changeset reverts the changed behavior. In order to preserve the
performance improvements introduced in [34561], an additional query must be
performed when 'default_comments_page=newest' and 'cpage=0' (ie, you're viewing
the post permalink). A nice side effect of this revert is that we no longer
need the hacks required to determine proper comment pagination, introduced in
[34561].
See #8071. See #34073.
git-svn-id: https://develop.svn.wordpress.org/trunk@34729 602fd350-edb4-49c9-b593-d223f7449a82
Continuing with the "code is poetry" theme after [34719], we need to continue
to update the option in the database on non-multisite in this
`wp_insert_term()` test.
See #31130.
git-svn-id: https://develop.svn.wordpress.org/trunk@34720 602fd350-edb4-49c9-b593-d223f7449a82
[34718] introduced a 'db_version' check to term meta functions, to ensure that
they don't run when the term meta schema is not yet in place. This call to
`get_option()` causes a database hit during Multisite tests, due to the
presence of the `WP_INSTALLING` constant. See #31130. The extra database
queries are causing cache tests to fail.
In similar cases, we have `markTestSkipped()` when `is_multisite()`. Because
the term meta API is so extensive - term meta caches can be primed anywhere a
`WP_Query` loop is fired up - we implement a more generous workaround in this
case. To prevent `get_option( 'db_version' )` from hitting the database during
multisite unit tests, we use a 'pre_option_' filter.
Heaven help us.
See #34091.
git-svn-id: https://develop.svn.wordpress.org/trunk@34719 602fd350-edb4-49c9-b593-d223f7449a82
The tests included in [34669] contained a couple of problems:
* Comments were not always created in the expected order, due to the incorrect use of 'comment_date_gmt'.
* 'asc'/'desc' and 'older'/'newer' were confused.
* 'default_comments_page=newest' ('last') didn't properly account for the cpage=1 offset.
The code itself powering this pagination was correct; it's only the tests that
were wrong.
See #8071.
git-svn-id: https://develop.svn.wordpress.org/trunk@34713 602fd350-edb4-49c9-b593-d223f7449a82
Lazy-loading logic is moved to a method on `WP_Query`. This makes it possible
for comment feeds to take advantage of metadata lazyloading, in addition to
comments loaded via `comments_template()`.
This new technique parallels the termmeta lazyloading technique introduced in
[34704].
Fixes#34047.
git-svn-id: https://develop.svn.wordpress.org/trunk@34711 602fd350-edb4-49c9-b593-d223f7449a82
[34529] introduced lazyloading for the metadata belonging to terms matching
posts in the main `WP_Query`. The current changeset improves this technique
in the following ways:
* Term meta lazyloading is now performed on the results of all `WP_Query` queries, not just the main query.
* Fewer global variable touches and greater encapsulation.
* The logic for looping through posts to identify terms is now only performed once per `WP_Query`.
Props dlh, boonebgorges.
See #34047.
git-svn-id: https://develop.svn.wordpress.org/trunk@34704 602fd350-edb4-49c9-b593-d223f7449a82
The 'exclude_tree' parameter must be compatible with 'hierarchical';
previously, 'hierarchical' canceled it out. This changeset also makes it so
that 'exclude_tree' is compatible with 'exclude'. When both are passed, and
'hierarchical' is true, the descendant trees of terms in both parameters will
be excluded from matched terms.
Props tott, webord, MikeHansenMe.
Fixes#12981.
git-svn-id: https://develop.svn.wordpress.org/trunk@34696 602fd350-edb4-49c9-b593-d223f7449a82
Gary Pendergast
Aaron Jorbin
Scott Taylor
Weston Ruter
Boone Gorges
Drew Jaynes
Sergey Biryukov
John Blackbourn
Jeremy Felt
Dion Hulse