03: First green CI - #3
Open
nikolaystrikhar wants to merge 9 commits into
Open
Conversation
Five defects from the PR review, all latent because tests/unit was empty:
- tests/unit/_bootstrap.php was missing while unit.suite.yml declares
`bootstrap: _bootstrap.php`. Codeception resolves that against the suite
path and throws ConfigurationException("Bootstrap file ... can't be
loaded") before any test runs. The file is docblock-only: the global
tests/_bootstrap.php already loads the autoloader.
- codeception.dist.yml had no `namespace`, so `codecept build` generated
the actions trait as `_generated\UnitTesterActions` at the root, while
UnitTester.php's trait use resolves relative to its own namespace. The
actor could never be instantiated, and the generated file was not
autoloadable under the Tests\Support PSR-4 mapping either.
- Dropped the root `actor: Tester`. The only suite declares its own actor,
and combined with `actor_suffix: Tester` the key implied "TesterTester".
- Dropped `theme: twentytwentythree`. It was pinned against
WP_VERSION=latest, and core rotates bundled default themes; nothing in
the suite needs a specific theme.
- The multisite env now uses its own table prefix. Both envs share a
database and WPLoader drops and recreates every table under its prefix
on install, so a shared prefix makes concurrent env runs race.
Also adds tests/_data/.gitkeep, since codeception.dist.yml declares the
path but the directory was never created.
Verified: `codecept build` now emits
Nexcess\PluginAbsorber\Tests\Support\_generated, the actor class resolves
with the trait applied, the suite bootstrap loads, and the multisite env
merges to multisite=true / prefix=mstest_.
slic reads SSH_AUTH_SOCK unconditionally when preparing its docker-compose env, and GitHub Actions runners don't have an ssh-agent running by default, so every matrix leg failed at 'slic here' with 'SSH_AUTH_SOCK environment variable is not set!'.
d4mation
reviewed
Jul 31, 2026
Drops tests/_support/Traits/WithUopz.php. lucatume\WPBrowser\Traits\UopzFunctions ships with wp-browser, is maintained by its author, undoes overrides through its own @after, and exists as far back as the ^3.6.5 floor this library pins, so there is no copy left to drift against the other plugin repos. Its setFunctionReturn() also takes an explicit $execute flag rather than inferring one from the value. exit is never mocked. Neutralising exit lets a test keep running past the point production would have halted, so a test that should fail can report as passing and CI will not say otherwise. test_exit_can_be_neutralised is gone; redirect branches are tested by stubbing the call immediately before exit and throwing TestException from it, with a worked example in tests/README.md. CI: narrow the token to contents:read, cancel superseded PR runs, add a 20 minute timeout, upload tests/_output on failure, and run multisite even when singlesite failed so one run reports both. The matrix is PHP 7.4 and 8.5 against WordPress latest and nightly -- four legs, nightly non-blocking. Every leg now pins core explicitly: the slic image bakes in a WordPress that varies by PHP version, so the previous "latest" legs silently tested whatever core the image happened to ship rather than latest. composer.json maps Tests\Unit\ to tests/unit. The previous Tests\ => tests/ mapping resolved SmokeTest to tests/Unit/SmokeTest.php, which Composer skips from the optimised classmap on a case-sensitive filesystem. Plan and design docs updated so Tasks 8 and 13 do not reintroduce the trait. Verified locally: slic run unit --env singlesite and --env multisite both report OK (3 tests, 5 assertions), the PSR-4 warning is gone, and the multisite env really installs a network (mstest_site and mstest_sitemeta present). Claude-Session: https://claude.ai/code/session_01YWju62NPWe1dtp1fysao9Q
nikolaystrikhar
force-pushed
the
03-ci-tests
branch
from
July 31, 2026 15:47
33c50c3 to
6a67f2d
Compare
# Conflicts: # composer.json
Codeception aborts with ConfigurationException("register_argc_argv must be
set to On") before it builds. slic's php.ini does not set the directive, so
the base image default applies and it differs by PHP version: the 7.4 legs
passed, both 8.5 legs failed at 'slic cc build'.
slic is checked out into the workspace and its containers/slic/php.ini is
bind-mounted as conf.d/zz-docker.ini, which loads after the main php.ini, so
appending the directive there overrides the image default.
Claude-Session: https://claude.ai/code/session_01YWju62NPWe1dtp1fysao9Q
The rule that exit is never mocked -- stub the call before it and throw TestException -- shipped documented in three places and exercised by zero tests. TestException had no callers at all. The smoke test exists to prove the harness assumptions before library code leans on them, and it proved two of them (WordPress loads, a function can be stubbed) while the newest and least conventional one went unchecked: that an exception raised inside a uopz-replaced function escapes it and is catchable by the test. test_a_stub_can_throw_to_halt_a_code_path closes that. Verified the mechanism independently of CI first, in a php:8.3-cli container with uopz built from pecl: uopz_set_return( $fn, $closure, true ) with a throwing closure propagates the exception out of the replaced function with its message intact. test_the_env_matches_its_table_prefix gives the multisite CI leg something to prove. Both envs previously ran the identical three assertions, so a multisite env that quietly failed to install a network -- or that picked up the wrong tables -- still reported green. Asserting is_multisite() against base_prefix covers the network and the per-env tablePrefix together. tests/README.md: the worked example did not compile ($subject was never defined) and stacked expectException, a try/catch and a re-throw to do one job. It is now the catch-and-flag shape, with the reason spelled out -- catching without asserting the exception arrived turns "the code never redirected" into a silent pass, which is the failure mode the section opens by warning about. Claude-Session: https://claude.ai/code/session_01YWju62NPWe1dtp1fysao9Q
Five things that were each fine while nothing moved. slic was checked out at ref: main, so an upstream reorganisation could change CI without a commit here -- and the register_argc_argv step reaches directly into slic's file layout. Pinned to 2.3.0, which is the same commit main points at today, so behaviour is unchanged. That step also used a bare >> redirect, which creates the file if slic ever moves or renames containers/slic/php.ini. The override would stop applying silently and surface later as an unrelated-looking cc build failure on 8.5 only. It now fails loudly when the target is missing. The multisite step ran on !cancelled(), whose stated intent was "run even when singlesite failed". It also ran after a failed composer install or build, stacking a second misleading failure on the real one. Now gated on the build having succeeded. continue-on-error keeps a red nightly leg from blocking the PR, but it also makes a failing leg report as a pass in the checks list -- the only way to notice was to open the run. The leg stays non-blocking; a failure now leaves a warning annotation and a job-summary note where someone will see it. Dropping continue-on-error entirely is a one-line change if a plain red X is preferred; nothing is gated on it today, since the repository has no branch protection or rulesets. WP_VERSION=latest in .env.testing.slic was inert -- the workflow's own comment says so, which is why the site-cli core update step exists -- and on the nightly legs it claimed the wrong thing. Removed, with a note pointing at what actually sets the version. Also: workflow_dispatch, so a nightly leg can be re-run without pushing; the checkout directory is no longer hardcoded; and the redundant CI=1 export is gone, since Actions already sets it. Claude-Session: https://claude.ai/code/session_01YWju62NPWe1dtp1fysao9Q
nikolaystrikhar
marked this pull request as draft
July 31, 2026 16:13
The new smoke test failed on all four legs with [Error] Cannot access self:: when no class scope is active which is the test doing its job on its first run. uopz executes a replacement closure outside the test object, and my local check missed it because calling a static closure directly from PHP does keep class scope -- only the uopz path does not. Probed the actual boundary in a container running uopz. Inside a closure installed with setFunctionReturn( $fn, $closure, true ): $this fatal, "Using $this when not in object context" self::CONST fatal, in static and non-static closures alike use ( ... ) works, including by reference arguments arrive normally So the message is resolved into a local and captured by value. This matters well beyond the smoke test. The planned Resolver and end-to-end tests stub deactivate_plugins() and wp_safe_redirect() with closures that collect into $this->deactivations / $this->redirects and throw with self::HALTED_AT_EXIT -- every one of those would have failed the same way, in a later PR, with nothing to point at the cause. The working shape is to bind a reference to the property, resolve constants into locals and capture both: $deactivations = &$this->deactivations; $halt_message = self::HALTED_AT_EXIT; Verified through uopz that writes made through the captured reference do land on the property. tests/README.md now documents the constraint with both the broken and the working form. Also drops a reference to WP_VERSION from a workflow comment, since the key it named no longer exists. Claude-Session: https://claude.ai/code/session_01YWju62NPWe1dtp1fysao9Q
nikolaystrikhar
marked this pull request as ready for review
July 31, 2026 16:18
d4mation
approved these changes
Jul 31, 2026
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Smoke test proving the harness, plus the PHP tests workflow.
Uses wp-browser's
UopzFunctionsinstead of a localWithUopztrait.