While the `skipTestOnTimeout()` method will catch a timeout and prevent it from causing a test to fail, other errors such as a failed DNS lookup or HTTPS handshake can still cause a test to unnecessarily fail. This introduces a simple retry mechanism that will hopefully further reduce the flakiness of tests that perform HTTP API requests.
Fixes#62830
git-svn-id: https://develop.svn.wordpress.org/trunk@59729 602fd350-edb4-49c9-b593-d223f7449a82
Introduces `WP_UnitTestCase_Base::assertSamePathIgnoringDirectorySeparators()` and an associated helper method `WP_UnitTestCase_Base::normalizeDirectorySeparatorsInPath()` to allow for comparing two file path strings independently of OS-specific differences.
The normalization is done in a separate method to also allow this method to be used for path normalization within test methods themselves, like for normalizing a group of paths in an array.
The pretty specific method name for the helper (`normalizeDirectorySeparatorsInPath()`) is an attempt to prevent naming conflicts with methods which may exist in plugin test suites build on top of the WP Core test suite.
Props jrf, hellofromTonya.
See #61530.
git-svn-id: https://develop.svn.wordpress.org/trunk@59057 602fd350-edb4-49c9-b593-d223f7449a82
This fixes a bug introduced by [57129] and [56635] in which deprecating the previous `TEMPLATEPATH` and `STYLESHEETPATH` constants in favor of `get_template_directory()` and `get_stylesheet_directory()` functions caused the active theme template path to change when using `switch_to_blog()`.
This introduces a new function, `wp_set_template_globals()`, which is called during the bootstrap process to store the template paths to new globals values `$wp_template_path` and `$wp_stylesheet_path`. This restores behavior to how things worked prior to [56635] but retains the ability for template values to be reset for better testability.
Related #18298, #60025.
Props joemcgill, flixos90, mukesh27, swissspidy, manfcarlo, metropolis_john, jeremyfelt.
Fixes#60290.
git-svn-id: https://develop.svn.wordpress.org/trunk@57685 602fd350-edb4-49c9-b593-d223f7449a82
Use 'theme_files' cache group for block pattern caches. Previously, block pattern cache data was not stored in a cache group and used the default group. This new cache group, is setup as a global cache group, meaning that sites using multisite, will have a single cache for block pattern data per theme. This change also no longer invalidate block pattern caches in multisite instances, meaning block pattern caches can be shared between sites on a network, meaning less repeated data in the object cache.
Props spacedmonkey, flixos90, joemcgill.
Fixes#60120.
git-svn-id: https://develop.svn.wordpress.org/trunk@57608 602fd350-edb4-49c9-b593-d223f7449a82
This allows third-party plugins to write their own factory extending `WP_UnitTest_Factory` for testing purposes, as well as benefit from `WP_UnitTestCase_Base` features.
Follow-up to [35186], [35225], [35242].
Props hugod.
Fixes#59999.
git-svn-id: https://develop.svn.wordpress.org/trunk@57149 602fd350-edb4-49c9-b593-d223f7449a82
Similar to other function in the `_deprecated_*` series, `_deprecated_class()` comes with two new hooks: `deprecated_class_run` and `deprecated_class_trigger_error`.
Support has also been added for setting class deprecation expectations in tests.
Props jrf, wvega, ohryan.
See #41125.
git-svn-id: https://develop.svn.wordpress.org/trunk@56467 602fd350-edb4-49c9-b593-d223f7449a82
[https://github.com/sebastianbergmann/phpunit/blob/9.6/ChangeLog-9.6.md#961---2023-02-03 PHPUnit 9.6.1] deprecated the `assertObjectHasAttribute()` and `assertObjectNotHasAttribute()` methods, leading to deprecation notices in a number of tests.
[https://github.com/sebastianbergmann/phpunit/blob/10.1.3/ChangeLog-10.1.md#1010---2023-04-14 PHPUnit 10.1.0] brought the methods back by popular request, though renamed as `assertObjectHasProperty()` and `assertObjectNotHasProperty()`, to prevent confusion with PHP 8.0 attributes.
This meant that users which cannot (yet) upgrade to PHPUnit 10.1+ would always have deprecation notices for these methods without recourse. So, after much discussion, the new methods have been backported to [https://github.com/sebastianbergmann/phpunit/blob/9.6/ChangeLog-9.6.md#9611---2023-08-19 PHPUnit 9.6.11], leaving just the 10.0.x series with a deprecation notice and no recourse.
What does this mean for WordPress?
WordPress uses the [https://github.com/Yoast/PHPUnit-Polyfills PHPUnit Polyfills] to be able to write tests for the most recent versions of PHPUnit, with the Polyfills taking care of polyfilling any new PHPUnit methods on older PHPUnit versions.
* The PHPUnit Polyfills 1.x series supports PHPUnit 4.x to 9.x.
* The PHPUnit Polyfills 2.x series supports PHPUnit 5.x to 10.x.
WordPress currently runs against PHPUnit 6.x to 9.x with PHPUnit Polyfills 1.x, while the new methods were previously only included in PHPUnit Polyfills 2.0.0+, as they were introduced in PHPUnit 10.x.
Since the `assertObjectHasProperty()` and `assertObjectNotHasProperty()` methods have been backported to PHPUnit 9.x, the PHPUnit Polyfills will now include these methods in the 1.x series as well.
By upgrading to the latest [https://github.com/Yoast/PHPUnit-Polyfills/releases/tag/1.1.0 PHPUnit Polyfills 1.1.0] release, we can get rid of the deprecation notices related to the use of the `assertObjectHasAttribute()` and `assertObjectNotHasAttribute()` methods.
This could have implications for plugins or themes running integration tests with WordPress if they have set their PHPUnit Polyfills dependency to a fixed version or have a too strict version constraint (limiting the PHPUnit Polyfills to the 1.0.x series). The solution for those plugins or themes is to update their version constraints for the PHPUnit Polyfills to allow for the 1.1.x series.
Follow-up to [51559], [51598].
Props jrf, ayeshrajans.
Fixes#59150.
git-svn-id: https://develop.svn.wordpress.org/trunk@56421 602fd350-edb4-49c9-b593-d223f7449a82
This resets the main query variable, `$wp_the_query` during the `WP_UnitTestCase_Base::tear_down` method that runs after each PHPUnit test. This ensures any changes to the main query object is reset after each test to avoid cross contamination between tests, similar to how other globals are reset.
Props flixos90, spacedmonkey, joemcgill.
Fixes#58776.
git-svn-id: https://develop.svn.wordpress.org/trunk@56212 602fd350-edb4-49c9-b593-d223f7449a82
The `usermeta` group was introduced in WP 2.0, but was never actually used in WordPress core.
The `user_meta` group is used instead.
Follow-up to [3011], [15482], [16562], [53823], [53832], [54940].
Props ignatggeorgiev, peterwilsoncc, oglekler, pamprn09, spacedmonkey, SergeyBiryukov.
Fixes#58175.
git-svn-id: https://develop.svn.wordpress.org/trunk@55940 602fd350-edb4-49c9-b593-d223f7449a82
In [36566] a framework to lazily load metadata was introduced. This supported term and comment meta by default. In this commit, extends support for site ( blog ) meta. Site meta is not heavily used by core and is used by developers to extend multisite. In this change, `_prime_site_caches` and `WP_Site_Query` now call the new function `wp_lazyload_site_meta`. The function `wp_lazyload_site_meta` accepts an array of ids and adds them to the queue of metadata to be lazily loaded. The function `get_blogs_of_user` was updated to now lazily load site meta.
Follow on from [55671].
Props spacedmonkey, johnjamesjacoby, peterwilsoncc, mukesh27.
Fixes#58185.
git-svn-id: https://develop.svn.wordpress.org/trunk@55747 602fd350-edb4-49c9-b593-d223f7449a82
In `WP_UnitTestCase::flush_cache` method, the properties of the global `$wp_object_cache` object were manaully being reset to flush the cache. The function `wp_cache_flush_runtime` was added in [52772] and is designed to reset any class properties to default values. Using `wp_cache_flush_runtime` improve compatibility with third party object caches, as it allows developers to define their own `wp_cache_flush_runtime` function.
Props rmccue, johnbillion, wonderboymusic, boonebgorges, voldemortensen, dd32, DrewAPicture, tillkruess, spacedmonkey.
Fixes#31463.
git-svn-id: https://develop.svn.wordpress.org/trunk@55741 602fd350-edb4-49c9-b593-d223f7449a82
The cache group `users-queries` was added in [55657]. This global cache group, was named to be inline with cache groups added in [55526]. However, the naming of the group does not match, as other cache groups, do not end with s at the end. This change fix this naming.
Props spacedmonkey, SergeyBiryukov, peterwilsoncc.
See #40613.
git-svn-id: https://develop.svn.wordpress.org/trunk@55680 602fd350-edb4-49c9-b593-d223f7449a82
In [34529] introduced lazy loading of term meta. However, this was only in the context of `WP_Query`. Other parts of the codebase, like `WP_Term_Query` did not lazily load term meta. In this change, calls to `update_termmeta_cache` are now replaced with `wp_lazyload_term_meta`, that instead of priming term meta caches, just adds them to the queue to be primed it ever called. This results in far less database queries, as there a number of places where term meta is being primed unnecessarily and never used. Adding everything to the term meta queue, also means that if term meta is used, that is all loaded in a single database / cache call.
Props spacedmonkey, mukesh27, peterwilsoncc.
Fixes#57645.
git-svn-id: https://develop.svn.wordpress.org/trunk@55671 602fd350-edb4-49c9-b593-d223f7449a82
Cache the results of database queries within `WP_User_Query` class. Only cache queries that are requesting 3 or less fields so that caches are not storing full user objects. Cache results are stored in a new global cache group named `users-queries`. Add a new parameter to `WP_User_Query` called `cache_results` to allow developers to opt out of a receiving cached results. `cache_results` parameter defaults to true. Also add a new helper function called `wp_cache_set_users_last_changed`, similar to `wp_cache_set_posts_last_changed` that incroments last changed value in cache group `users`. Ensure that `wp_cache_set_users_last_changed` is called whenever user / user meta is modified for proper cache invalidation.
Props johnjamesjacoby, spacedmonkey, westi, dd32, strategio, srikanthmeenakshi, OllieJones, khoipro, rjasdfiii, flixos90, mukesh27, peterwilsoncc.
Fixes#40613.
git-svn-id: https://develop.svn.wordpress.org/trunk@55657 602fd350-edb4-49c9-b593-d223f7449a82
Fixes a bug introduced in [55526]. The cache groups `network-queries` and `site-queries` store query results for network and site. These are network level data and should be stored in a global cache group.
Props spacedmonkey, tillkruess, flixos90.
See #57625.
git-svn-id: https://develop.svn.wordpress.org/trunk@55537 602fd350-edb4-49c9-b593-d223f7449a82
In [47938] the `wp_cache_get_multiple` function was added to core. This function allows for multiple cache keys to be received from cache in a single function call. `wp_queue_posts_for_term_meta_lazyload` function does many calls to cache. To get taxonomy relationship for multiple posts and get all terms. Replace calls to `get_object_term_cache` with calls to `wp_cache_get_multiple` and `_prime_term_caches`. This improves performance on sites that implement the `wp_cache_get_multiple` in their object caching drop-in.
Props spacedmonkey, ocean90, SergeyBiryukov, costdev, flixos90, joemcgill, 10upsimon.
Fixes#57150.
git-svn-id: https://develop.svn.wordpress.org/trunk@55252 602fd350-edb4-49c9-b593-d223f7449a82
Introduced along with the `did_filter()` function, the `$wp_filters` global stores the number of times each filter has been applied during the current request.
This commit adds `$wp_filters` to the list of globals that are saved in `WP_UnitTestCase_Base::_backup_hooks()` and restored in `::_restore_hooks()` so that hooks set by the current test do not accidentally affect future tests.
Additionally, this commit brings some consistency by backing up and restoring the hook-related globals in the same order they are defined in `wp-includes/plugin.php`.
Follow-up to [29251], [53803].
Props sanjucta, SergeyBiryukov.
Fixes#57236.
git-svn-id: https://develop.svn.wordpress.org/trunk@55160 602fd350-edb4-49c9-b593-d223f7449a82
This changeset is part of a greater effort to enhance the caching strategy for `theme.json` based data. Similar to [55138], the cache is currently ignored when `WP_DEBUG` is on to avoid interrupting the theme developer's workflow.
Props oandregal, spacedmonkey, hellofromtonya, flixos90, ironprogrammer, azaozz, aristath, costdev, mcsf.
Fixes#56910.
git-svn-id: https://develop.svn.wordpress.org/trunk@55148 602fd350-edb4-49c9-b593-d223f7449a82
Calls to `wp-admin/load-styles.php` do not include the loading of `wp_cache_*()` functions. With [55086], this caused a fatal error:
{{{
Fatal error: Uncaught Error: Call to undefined function wp_cache_get() in /wp-includes/global-styles-and-settings.php on line 285
}}}
In some production and local environments running `trunk`, the admin area looked broken as the styling was not loaded as there were no HTTP requests.
This commit reverts the caching from [55086] to restore sites running `trunk` until a solution is found.
Follow-up to [55086].
Props Otto42, dmsnell, costdev.
See #56975.
git-svn-id: https://develop.svn.wordpress.org/trunk@55092 602fd350-edb4-49c9-b593-d223f7449a82
Adds `wp_theme_has_theme_json()` for public consumption, to replace the private internal Core-only `WP_Theme_JSON_Resolver::theme_has_support()` method. This new global function checks if a theme or its parent has a `theme.json` file.
For performance, results are cached as an integer `1` or `0` in the `'theme_json'` group with `'wp_theme_has_theme_json'` key. This is a non-persistent cache. Why? To make the derived data from `theme.json` is always fresh from the potential modifications done via hooks that can use dynamic data (modify the stylesheet depending on some option, settings depending on user permissions, etc.).
Also adds a new public function `wp_clean_theme_json_cache()` to clear the cache on `'switch_theme'` and `start_previewing_theme'`.
References:
* [https://github.com/WordPress/gutenberg/pull/45168 Gutenberg PR 45168] Add `wp_theme_has_theme_json` as a public API to know whether a theme has a `theme.json`.
* [https://github.com/WordPress/gutenberg/pull/45380 Gutenberg PR 45380] Deprecate `WP_Theme_JSON_Resolver:theme_has_support()`.
* [https://github.com/WordPress/gutenberg/pull/46150 Gutenberg PR 46150] Make `theme.json` object caches non-persistent.
* [https://github.com/WordPress/gutenberg/pull/45979 Gutenberg PR 45979] Don't check if constants set by `wp_initial_constants()` are defined.
* [https://github.com/WordPress/gutenberg/pull/45950 Gutenberg PR 45950] Cleaner logic in `wp_theme_has_theme_json`.
Follow-up to [54493], [53282], [52744], [52049], [50959].
Props oandregal, afragen, alexstine, aristath, azaozz, costdev, flixos90, hellofromTonya, mamaduka, mcsf, ocean90, spacedmonkey.
Fixes#56975.
git-svn-id: https://develop.svn.wordpress.org/trunk@55086 602fd350-edb4-49c9-b593-d223f7449a82
In various unit test factory classes, some of the `create_object()` and `update_object()` methods returned a `WP_Error` object on failure, while a few others were documented to do so, but did not in practice, instead returning the value `0` or `false`, or not accounting for a failure at all.
This commit aims to handle this in a consistent way by updating the methods to always return the object ID on success and a `WP_Error` object on failure.
Includes:
* Updating and correcting the relevant documentation parts.
* Adding missing documentation and `@since` tags in some classes.
* Renaming some variables to clarify that it is the object ID which is passed around, not the object itself.
Follow-up to [760/tests], [838/tests], [922/tests], [948/tests], [985/tests], [27178], [32659], [34855], [37563], [40968], [44497], [46262].
See #56793.
git-svn-id: https://develop.svn.wordpress.org/trunk@55019 602fd350-edb4-49c9-b593-d223f7449a82
The `setUp()` and `tearDown()` methods were renamed to `set_up()` and `tear_down()`, respectively, as part of implementing the `void` return type solution for PHPUnit 8.0.
Follow-up to [29120], [29251], [30277], [32173], [32806], [38829], [42379], [50450], [51276], [51568].
See #56793.
git-svn-id: https://develop.svn.wordpress.org/trunk@54865 602fd350-edb4-49c9-b593-d223f7449a82
As of [35186] and [51568], there are two sets of methods used for setup/teardown in the test suite before and after a test class is run:
* `set_up_before_class()` / `tear_down_after_class()`
* `wpSetUpBeforeClass()` / `wpTearDownAfterClass()`. (Note the `wp` prefix, these are WordPress' own methods and are not the same as the native PHPUnit `setUpBeforeClass()` / `tearDownAfterClass()` methods.)
The main difference is that `wpSetUpBeforeClass()` receives the `$factory` argument for ease of use, and both `wpSetUpBeforeClass()` and `wpTearDownAfterClass()` don't need to call `self::commit_transaction()`.
Many tests use the `wpTearDownAfterClass()` method to clean up posts, users, roles, etc. created via `wpSetUpBeforeClass()`. However, due to [source:tags/6.0/tests/phpunit/includes/abstract-testcase.php?marks=88-95#L82 how the method was previously called], this cleanup happened after all data is **already deleted** from the database.
This could cause some confusion when refactoring tests. For example:
{{{
public static function wpTearDownAfterClass() {
$GLOBALS['_wp_additional_image_sizes'] = self::$_sizes;
}
public static function tear_down_after_class() {
wp_delete_attachment( self::$large_id, true );
parent::tear_down_after_class();
}
}}}
At a glance, it seems like these two methods can be combined:
{{{
public static function wpTearDownAfterClass() {
wp_delete_attachment( self::$large_id, true );
$GLOBALS['_wp_additional_image_sizes'] = self::$_sizes;
}
}}}
However, that would not work as expected: by the time `wp_delete_attachment()` runs, the attachment ID is no longer in the database, so it returns early, leaving some files in the `uploads` directory.
By calling `wpTearDownAfterClass()` in `WP_UnitTestCase_Base::tear_down_after_class()` before deleting all data, instead of after, we ensure that both of these methods have access to the same data and can be used interchangeably to perform cleanup as necessary.
Additionally, this commit moves the calls to parent methods in `WP_UnitTestCase_Base`:
* `parent::set_up_before_class()` to be the first thing called in `::set_up_before_class()`
* `parent::tear_down_after_class()` to be the last thing called in `::tear_down_after_class()`
This does not have any effect in practice, but brings consistency with how these methods are called in the test suite.
Follow-up to [35186], [35225], [35242], [38398], [39626], [49001], [51568].
Props ironprogrammer, SergeyBiryukov.
Fixes#55918. See #55652.
git-svn-id: https://develop.svn.wordpress.org/trunk@54366 602fd350-edb4-49c9-b593-d223f7449a82
This appears to need more investigation. Instead, delete the `test-plugin` and `link-manager` directories in REST API plugins controller tests, for which this change was initially intended.
Follow-up to [54300], [54301], [54303].
See #55652, #56629.
git-svn-id: https://develop.svn.wordpress.org/trunk@54304 602fd350-edb4-49c9-b593-d223f7449a82
Includes:
* Checking if the directory exists and returning early otherwise.
* Removing a redundant `rmdir()` call in `clean_dirsize_cache()` tests.
Follow-up to [49744], [54300].
See #55652.
git-svn-id: https://develop.svn.wordpress.org/trunk@54303 602fd350-edb4-49c9-b593-d223f7449a82
The `WP_UnitTestCase_Base::rmdir()` method selectively deletes files from a directory, skipping any paths from the `$ignore_files` property.
This commit updates the method to remove the empty directory if there are no files left, bringing some parity with PHP native `rmdir()` function.
Follow-up to [677/tests], [29120].
See #55652.
git-svn-id: https://develop.svn.wordpress.org/trunk@54300 602fd350-edb4-49c9-b593-d223f7449a82
These magic methods were introduced to prevent a backward compatibility break, but in actual fact:
1. ''Caused'' a backward compatibility break. The original `$factory` property was a `static` property and this declared property was replaced by the magic methods. Unfortunately, it was not realized at the time that these magic methods **''are not called for static property access''**.[[BR]][[BR]]
> Property overloading only works in object context. These magic methods will not be triggered in static context.
And as approaching a static property in a non-static manner is [https://3v4l.org/93HQL not supported in PHP], this effectively created a backward compatibility break instead of preventing it.
2. Were hiding errors in tests, as the magic methods would be invoked for non-existent properties and would return `null` (get) or `false` (isset). See [54040], [54041], and [54077] for bug fixes related to this.
3. Are problematic in relation to PHP 8.2, as the implementation is incomplete, does not protect against dynamic properties and hides PHP notices about undefined properties.
Now, there were several options to mitigate this:
1. Revert the original commit. This would be problematic, as the ''non-static'' version of these properties has now been supported for 7 years, so this would create a new backward compatibility break.
2. Improve the magic methods. With all the issues with magic methods (see the discussion in the [https://www.youtube.com/watch?v=vDZWepDQQVE livestream from August 16, 2022], this would probably cause more problems than it’s worth and would make for a much more complex implementation, which is over the top for this relatively simple functionality, especially in the context of a test suite.
3. Remove the magic methods without adding the property. This would again cause a backward compatibility break, though one for which the mitigation solution would be relatively straightforward, i.e. to replace property access using `$this->factory` with a function call `$this->factory()` (or `self::factory()`, as the method is declared as static). While we can (and have in a subsequent commit) mitigate this for the WP Core test suite, mitigating this for plugin or theme integration tests is outside of our purview and they would still need to deal with this backward compatibility break.
4. The current solution: removing the magic methods, explicitly declaring the (non-static) property and setting it in the `set_up()` method. This does not constitute a backward compatibility break with the functionality as it was over the past 7 years. Setting the property in `set_up()` may be “late”, but that is the earliest place in which the property can be set as non-static. If the factory would be needed prior to `set_up()`, the (static) `WP_UnitTestCase_Base::factory()` method should be called directly. This is no different from how this functionality behaved over the past 7 years.
Note: The property is straight away marked as “deprecated”, since the method should be favored over the use of the property.
Reference: [https://www.php.net/manual/en/language.oop5.overloading.php#object.get PHP Manual: Property overloading: __get()]
Follow-up to [35225], [35242].
Props jrf, costdev.
See #56514.
git-svn-id: https://develop.svn.wordpress.org/trunk@54087 602fd350-edb4-49c9-b593-d223f7449a82
Dynamic (non-explicitly declared) properties are deprecated as of PHP 8.2 and are expected to become a fatal error in PHP 9.0.
In this particular case, the `scandir()` method sets a dynamic `$matched_dirs` property, which is subsequently used in the `delete_folders()` method.
This commit removes the need for the property. Effectively, this changes the `scandir()` method to return an array of the matched directories instead of setting the property by using recursion in the method itself in an optimized manner.
Note the `array_merge()` not being in the loop itself, but at the very end of the function. This is for performance reasons, see [https://github.com/dseguy/clearPHP/blob/master/rules/no-array_merge-in-loop.md clearPHP: No array_merge() In Loops] for a more detailed explanation of this.
It has been verified in detail that the actual results of the previous version of the method and this version match, even when the paths passed are more complex and have deeper nested subdirectories.
Follow-up to [30658], [53557], [53558], [53850], [53851], [53852], [53853], [53854], [53856], [53916], [53935], [53936].
Props jrf, johnbillion, markjaquith, SergeyBiryukov.
See #56033.
git-svn-id: https://develop.svn.wordpress.org/trunk@53937 602fd350-edb4-49c9-b593-d223f7449a82
This brings the list of persistent groups when running the test suite in line with core.
Follow-up to [1332/tests], [37613], [52976].
See #55652.
git-svn-id: https://develop.svn.wordpress.org/trunk@53828 602fd350-edb4-49c9-b593-d223f7449a82
This aims to provide better context and more details if an unexpected `_doing_it_wrong()` message or deprecation notice is encountered during a test run.
Previously, this would display a message like `Unexpected incorrect usage notice for [...]`, but without any further details, making it harder to track down the actual issue.
Follow-up to [25402], [25408], [25785], [37861], [51872].
See #55652.
git-svn-id: https://develop.svn.wordpress.org/trunk@53637 602fd350-edb4-49c9-b593-d223f7449a82
Previously, in case of failure, `WP_UnitTestCase_Base::assertNotWPError()` displayed the actual error message from the passed `WP_Error` object, but only if the `$message` parameter was empty.
This made the assertion less helpful, as the actual error message was lost in case there was a non-empty `$message` parameter passed to the method, as per the [https://make.wordpress.org/core/handbook/testing/automated-testing/writing-phpunit-tests/#using-assertions Writing PHP Tests] guidelines:
> All PHPUnit assertions, as well as all WordPress custom assertions, allow for a `$message` parameter to be passed. This message will be displayed when the assertion fails and can help immensely when debugging a test. This parameter should always be used if more than one assertion is used in a test method.
This commit ensures that the actual error message is always displayed, in addition to the passed `$message` parameter.
The same applies to `WP_UnitTestCase_Base::assertNotIXRError()`.
Follow-up to [34638], [40417].
Props jrf, SergeyBiryukov.
See #55652.
git-svn-id: https://develop.svn.wordpress.org/trunk@53536 602fd350-edb4-49c9-b593-d223f7449a82
This introduces a new test helper function which allows for turning a single-level array containing text strings into a data provider with named data sets, where the value of the data set will also be used as the name of the data set.
The function contains safeguards to ensure that it is only used with data compatible with this principle and will throw generic PHP exceptions when the data is incompatible. These type of exceptions will be displayed before the tests even start running and will stop the test run when they occur.
While generally speaking, all test cases ''should'' extend the base `WP_UnitTestCase_Base` class, this is still made a `public static` method to allow for a test, which by exception directly extends the PHPUnit base `TestCase` or the `PHPUnit_Adapter_TestCase`, to also be able to use this method.
Typical usage of this method:
{{{
public function data_provider_for_test_name() {
$array = array(
'value1',
'value2',
);
return $this->text_array_to_dataprovider( $array );
}
}}}
The returned result will look like:
{{{
array(
'value1' => array( 'value1' ),
'value2' => array( 'value2' ),
)
}}}
Props jrf, hellofromTonya, adamsilverstein.
See #55652.
git-svn-id: https://develop.svn.wordpress.org/trunk@53521 602fd350-edb4-49c9-b593-d223f7449a82
The odd / even class attribute global variables are causing issues in comments tests when a new test is added or an existing test is modified. To stabilize the odd / even comment tests, the comment global variables are added to the base test class' `tear_down()` using the same patterns as the other global resets. This change ensures each comment test starts at the same state. In doing so, the expected odd / even class attributes are no longer affected by previous tests, i.e. test leaks.
Follow-up to [53172].
Props hellofromTonya, zieladam, peterwilsoncc.
See #54725.
git-svn-id: https://develop.svn.wordpress.org/trunk@53430 602fd350-edb4-49c9-b593-d223f7449a82
This ensures that `WP_UnitTestCase::skipOnAutomatedBranches()` has access to the `GITHUB_REF` and `GITHUB_EVENT_NAME` variables, so that some tests can be skipped when appropriate.
Additionally, account for renaming the `master` branch to `trunk` in November 2021.
Follow-up to [40241], [46999], [49264], [49267], [51868].
Fixes#55668.
git-svn-id: https://develop.svn.wordpress.org/trunk@53349 602fd350-edb4-49c9-b593-d223f7449a82
This resolves an issue where setting up `$_REQUEST['post_type']` and not clearing it afterwards in `Tests_Admin_IncludesScreen::setup_block_editor_test()` started affecting a few `WP_Comments_List_Table` tests after [51997]. It also ensures a similar issue does not inadvertently happen in other tests.
Follow-up to [760/tests], [51997].
See #53363.
git-svn-id: https://develop.svn.wordpress.org/trunk@51999 602fd350-edb4-49c9-b593-d223f7449a82
Since PHPUnit 9.5.10 and 8.5.21, PHP deprecations are no longer converted to exceptions by default (`convertDeprecationsToExceptions="true"` can be configured to enable this).
Reference: [fac02620f6 Do not convert PHP deprecations to exceptions by default]; [https://github.com/sebastianbergmann/phpunit/blob/9.5.10/ChangeLog-9.5.md#9510---2021-09-25 PHPUnit 9.5.10 changelog].
With this change, the test for the `Services_JSON` compat class started failing:
{{{
There was 1 failure:
1) Tests_Compat_jsonEncodeDecode::test_json_encode_decode
Failed asserting that exception of type "PHPUnit\Framework\Error\Deprecated" is thrown.
}}}
This converts the native PHPUnit `::expectDeprecation()` method call in the test to a set of individual WP-specific `::setExpectedDeprecated()` method calls in order to not depend on PHPUnit behavior that is no longer the default.
Additionally, this commit includes support for catching deprecation notices from `_deprecated_file()` function calls to the `WP_UnitTestCase_Base::expectDeprecated()` method.
Follow-up to [46205], [46625], [48996], [51563], [51852], [51871].
Props jrf, netweb, SergeyBiryukov.
See #54183, #54029, #53363.
git-svn-id: https://develop.svn.wordpress.org/trunk@51872 602fd350-edb4-49c9-b593-d223f7449a82
Basically, the whole `assertSameIgnoreEOL()` assertion was fundamentally flawed. The assertion contends that it checks that the expected and actual values are of the same type and value, but the reality was very different.
* The function uses `map_deep()` to potentially handle all sorts of inputs.
* `map_deep()` handles arrays and objects with special casing, but will call the callback on everything else without further distinction.
* The callback used passes the expected/actual value on to the `str_replace()` function to remove potential new line differences.
* And the `str_replace()` function will - with a non-array input for the `$subject` - always return a string.
* The output of these calls to `map_deep()` will therefore have "normalized" _all properties_ in objects, _all values_ in arrays and _all non-object, non-array values_ to strings.
* And a call to `assertSame()` will therefore NEVER do a proper type check as the type of all input has already, unintentionally, been "normalized" to string.
Aside from this clear flaw in the design of the assertion, PHP 8.1 now exposes a further issue as a `null` value for an object property, an array value or a plain value, will now yield a ` str_replace(): Passing null to parameter #3 ($subject) of type array|string is deprecated` notice.
To fix both these issues, the fix in this PR ensures that the call to `str_replace()` will now only be made if the input is a text string.
All other values passed to the callback are left in their original type.
This ensures that a proper value AND type comparison can be done as well as prevents the PHP 8.1 deprecation notices.
Ref:
* https://developer.wordpress.org/reference/functions/map_deep/
* https://www.php.net/manual/en/function.str-replace.php
This commit:
- Fixes type-casting of non-string values to `string` (the flawed part of this assertion) by invoking `str_replace()` when the value is of string type.
- Fixes the PHP 8.1 `str_replace(): Passing null to parameter #3 ($subject) of type array|string is deprecated` deprecation notice.
- Micro-optimization: skips `map_deep()` when actual and/or expected are `null` (no need to process).
- Adjusts the method documentation for both this method and the `assertEqualsIgnoreEOL()` alias method to document that the `$expected` and `$actual` parameters can be of any type.
Follow-up to [48937], [51135], [51478].
Props jrf, hellofromTonya.
See #53363, #53635.
git-svn-id: https://develop.svn.wordpress.org/trunk@51831 602fd350-edb4-49c9-b593-d223f7449a82
The `assertDiscardWhitespace()` method uses `assertEquals()` under the hood, meaning that in reality any type of actual/expected value should be accepted by the function. Fixed the documentation to reflect that.
At the same time, only strings can contain whitespace differences. So the whitespace replacement should only be done when string values are passed.
This change (a) prevents potential `passing null to non-nullable` errors on PHP 8.1, if either of the inputs would turn out to be `null` and (b) increases tests stability.
Follow-up to [35003], [44902].
Props jrf.
See #53363.
git-svn-id: https://develop.svn.wordpress.org/trunk@51698 602fd350-edb4-49c9-b593-d223f7449a82
The following changes improve tests stability.
The `assertEqualFields()` method expects an object and a fields array as inputs and subsequently approaches the received parameters as such, but did not verify whether the received parameters are of the expected types.
Along the same lines, the `assertSameSets()`, `assertEqualSets()`, `assertSameSetsWithIndex()` and the `assertEqualSetsWithIndex()` methods all expect arrays for both the actual as well as the expected values and uses the array function `[k]sort()` on both, but never verified that the received inputs were actually arrays, which could lead to PHP errors on the sorting function calls.
Follow-up to [30687], [42343], [48937], [48939], [51480], [51481].
Props jrf.
See #53363.
git-svn-id: https://develop.svn.wordpress.org/trunk@51697 602fd350-edb4-49c9-b593-d223f7449a82
When a closure does not use `$this`, it can be made `static` for improved performance.
Static closures are supported in PHP since PHP 5.4.
Props jrf, hellofromTonya, swissspidy, SergeyBiryukov.
See #53359.
git-svn-id: https://develop.svn.wordpress.org/trunk@51657 602fd350-edb4-49c9-b593-d223f7449a82
Update the `wp_die_handler()` method to pass the response into `WPDieException` so that `expectExceptionCode()` calls work as expected.
Follow-up to [1221/tests], [3934], [4009], [12309].
Props pbearne, jrf, mukesh27.
Fixes#53882.
git-svn-id: https://develop.svn.wordpress.org/trunk@51608 602fd350-edb4-49c9-b593-d223f7449a82
