Update README to change chunk file suffix #108

Workflow file for this run

.github/workflows/playwright.yml at 79e1e20

	name: Playwright tests 🎭

	# The only browser-driven suite in CI — it replaced the Behat workflow, whose
	# environment provisioning steps (up through MySQL) this one inherits.
	#
	# There is deliberately no Chrome/Chromedriver/Selenium/Java/Xvfb stack here:
	# tests/playwright/features/actionInstall.feature installs Chamilo by clicking
	# through the legacy web installer at /main/install/index.php, so Playwright's
	# own browser handles both installing the app and testing it — one
	# browser-automation stack for the whole job, not two.
	on: [ push, pull_request ]

	jobs:
	build:
	name: PHP ${{ matrix.php-versions }} Test on ${{ matrix.operating-system }}
	runs-on: ubuntu-24.04

	strategy:
	fail-fast: false
	matrix:
	operating-system: [ ubuntu-latest ]
	# 8.4/8.5 temporarily disabled: focusing on getting the full suite green on
	# 8.3 first, then re-enabling the rest — that split also makes any future
	# failure easier to read as "version-specific" instead of muddying it with
	# whatever's still being worked out on 8.3 alone. Re-enable by uncommenting.
	php-versions: [ '8.3' ] # , '8.4', '8.5'
	steps:
	- name: Checkout
	uses: actions/checkout@v7

	# As of PHP 8.5, opcache is compiled into core (php-src RFC "Make OPcache a
	# non-optional part of PHP") and no longer ships as its own loadable
	# extension/package — there's no php8.5-opcache in ondrej/php, and
	# requesting the "opcache" extension by name would fail. Both spots below
	# that reference it (this apt package list, and the extensions: list two
	# steps down) skip it for 8.5 accordingly. Computed here in plain bash
	# rather than a GitHub Actions `${{ cond && a \|\| b }}` expression — a first
	# attempt at that pattern broke on the classic JS-ternary-via-&&/\|\|
	# pitfall (the "true" branch value was '', which is falsy, so `\|\|` fell
	# through to the fallback branch instead — i.e. it did the opposite of
	# what was intended, and 8.5 tried to install php8.5-opcache anyway,
	# failing the whole apt-get install command). Extend the version check if
	# a future PHP version keeps this behavior, since it isn't range-proofed.
	- name: Setup apache
	run: \|
	PHP_VERSION="${{ matrix.php-versions }}"
	OPCACHE_PACKAGE="php${PHP_VERSION}-opcache"
	if [ "$PHP_VERSION" = "8.5" ]; then
	OPCACHE_PACKAGE=""
	fi
	sudo apt-get install software-properties-common
	sudo add-apt-repository ppa:ondrej/php
	sudo apt-get update
	sudo apt-get install -y nodejs npm apache2 \
	php${PHP_VERSION} \
	php${PHP_VERSION}-common \
	php${PHP_VERSION}-cli \
	php${PHP_VERSION}-mysql \
	php${PHP_VERSION}-bcmath \
	php${PHP_VERSION}-curl \
	php${PHP_VERSION}-gd \
	php${PHP_VERSION}-intl \
	php${PHP_VERSION}-ldap \
	php${PHP_VERSION}-mbstring \
	$OPCACHE_PACKAGE \
	php${PHP_VERSION}-soap \
	php${PHP_VERSION}-xml \
	php${PHP_VERSION}-zip \
	libapache2-mod-php${PHP_VERSION}
	sudo a2enmod rewrite actions
	sudo chmod 777 -R $HOME

	- name: Setup PHP, with composer and extensions
	uses: shivammathur/setup-php@v2
	with:
	php-version: ${{ matrix.php-versions }}
	extensions: bcmath, ctype, curl, dom, gd, iconv, intl, json, ldap, mbstring, pdo_mysql, mysqli, soap, xml, zip${{ matrix.php-versions != '8.5' && ', opcache' \|\| '' }}
	ini-values: post_max_size=256M, max_execution_time=600, memory_limit=4096M, date.timezone=Europe/Paris, opcache.enable=1, opcache.memory_consumption=64M

	- name: Get composer cache directory
	id: composer-cache
	run: echo "dir=$(composer config cache-files-dir)" >> $GITHUB_OUTPUT

	- name: Install dependencies with composer
	run: composer install --no-progress

	- name: Install assets
	run: php bin/console assets:install

	- name: Install Yarn
	run: \|
	sudo npm i -g npm
	node --version
	npm --version
	sudo corepack enable
	yarn set version stable
	yarn --version

	- name: Install JS dependencies
	run: yarn install

	- name: Yarn run encore
	run: yarn run encore production

	- name: Start mysql service
	run: sudo /etc/init.d/mysql start

	- name: Install Playwright browsers
	run: yarn playwright install --with-deps chromium

	# Deliberately last, right before Apache is actually needed, not
	# right after `apt-get install apache2` back in "Setup apache".
	# A previous run showed 000-default.conf was still the stock
	# /var/www/html vhost by this point in the job, despite "Setup
	# apache" copying gh-apache into it and re-enabling the site
	# without error — something in between (most likely
	# shivammathur/setup-php@v2, which manages the Apache/PHP
	# integration) was resetting it. Applying the vhost swap here
	# instead, as the last write before first use, sidesteps that
	# regardless of which exact step was clobbering it.
	- name: Configure apache vhost
	run: \|
	sudo a2dissite 000-default.conf
	sudo cp -f ${{ github.workspace }}/.github/gh-apache /etc/apache2/sites-available/000-default.conf
	sudo a2ensite 000-default.conf
	sudo systemctl restart apache2

	- name: Check apache is up
	run: \|
	sudo chmod 777 -R ${{ github.workspace }}
	curl http://localhost/main/install/index.php

	- name: Install chamilo
	env:
	BASE_URL: http://localhost
	run: \|
	yarn test:playwright:install
	sed -i "s/APP_ENV='dev'/APP_ENV='prod'/g" .env
	sed -i "s/APP_DEBUG='1'/APP_DEBUG='0'/g" .env
	if grep -q "APP_ENV='prod'" .env; then
	printf 'framework:\n http_cache: true\n' > config/packages/prod/framework.yaml
	fi
	sudo php bin/console cache:clear
	sudo chmod -R 777 var

	# Seeds the fixed test accounts (mmosquera, acostea, ptook, etc.)
	# that the rest of the suite logs in as via "I am a teacher" /
	# "I am a student" / etc. — kept as its own step, run once,
	# separate from "Playwright tests" below, same reasoning as
	# keeping "Install chamilo" separate.
	- name: Seed test users
	env:
	BASE_URL: http://localhost
	run: yarn test:playwright:seed

	# Creates the "TEMP" course (which becomes cid=3 on this fresh
	# install) as a one-time setup fixture, same reasoning as "Seed
	# test users" above: course_user_registration.feature and every
	# remaining course-tool feature file (toolAgenda, toolForum,
	# toolLp, etc. — 14 in total as of this writing) hardcode cid=3,
	# assuming "TEMP" already exists. Left as its own step, run once,
	# BEFORE the main parallel "Playwright tests" step below, so
	# every other feature file can rely on it regardless of which
	# worker picks them up or what order they run in — a real CI
	# failure showed course_user_registration.feature's first
	# scenario reaching that course before course.feature's own
	# course-creation scenario (in a different file, different worker)
	# had even run.
	#
	# WHY cid=3 AND NOT cid=1 (changed 2026-08-19): the installer's
	# own DemoCoursesFixtures (fixture group "install", loaded by the
	# plain `doctrine:fixtures:load` this workflow's "Install chamilo"
	# step runs) now creates TWO demo courses before any test-suite
	# seeding happens at all — "AI Act" (cid=1) and "Using Chamilo"
	# (cid=2), added to the product on 2026-08-08/2026-08-11, i.e.
	# AFTER this Playwright suite was originally written against the
	# older "TEMP is the first course, so it gets id 1" assumption.
	# That assumption is now permanently false, so the whole suite was
	# migrated cid=1 -> cid=3 in one pass. Note this failed QUIETLY
	# before the migration: cid=1 is a real, working course, so
	# cid=1-based scenarios kept passing while silently exercising the
	# AI Act demo course instead of TEMP.
	#
	# This step must create EXACTLY ONE course for cid=3 to be
	# deterministic. `--grep "Create a course before testing"` used to
	# match TWO scenarios — course.feature's ("TEMP") and
	# toolExerciseTeacher.feature's ("EXTEACH") — so on a multi-worker
	# pool whichever finished first took id 3 and the other took id 4,
	# a genuine coin flip (observed both ways: TEMP won on real CI,
	# EXTEACH won locally at 43.7s vs 54.6s). Fixed at the source by
	# renaming that other scenario to "Create the exercise teacher
	# course before testing", so this grep no longer matches it and
	# EXTEACH is created inline as its own file's first scenario
	# instead (it is referenced only by course CODE, never by cid, so
	# its id is irrelevant — see that file's comment). The
	# --workers=1 in package.json's test:playwright:seed-course is
	# kept purely as a guard: if a future feature file ever adds a
	# scenario matching this grep again, single-worker + the config's
	# `fullyParallel: false` keeps creation order deterministic
	# (sorted by file path, so "course.feature" still goes first)
	# instead of silently reshuffling every cid=3 URL in the suite.
	- name: Seed test course
	env:
	BASE_URL: http://localhost
	run: yarn test:playwright:seed-course

	# Same reasoning as "Seed test course" above, for the OTHER course
	# fixture: sessionAccess.feature (new) selects this course by its
	# code, "TEMPPRIVATE", to attach to the sessions it creates.
	# course.feature's own "Create a private course before testing"
	# scenario is what creates it, but wasn't part of the dedicated
	# seed step (only "Create a course before testing"/TEMP was) — the
	# exact same cross-file race gotcha already fixed once for TEMP/
	# cid=1 would otherwise apply here too, just for a course "TEMPPRIVATE"
	# is looked up by code instead of by an assumed id.
	- name: Seed private course
	env:
	BASE_URL: http://localhost
	run: yarn test:playwright:seed-private-course

	# Real, confirmed root cause: GroupSettingsSchema.php's schema
	# default for 'allow_group_categories' is 'false'. toolGroup.
	# feature's entire category/group listing depends on it being
	# 'true' for the whole file's run. adminSettings.feature's own
	# "Update 'allow_group_categories' setting" scenario sets it to
	# 'Yes', but that Feature is tagged @settings and its BeforeAll/
	# AfterAll pair only restores it back to whatever was snapshotted
	# BEFORE that scenario ran — 'false' on a fresh install — once
	# its own scenarios finish. Since different feature files still
	# run concurrently across workers, toolGroup.feature only saw
	# 'true' during the narrow window while adminSettings.feature's
	# own scenario had it temporarily flipped — a real CI run showed
	# toolGroup.feature's very first scenario seeing genuinely zero
	# categories rendered, no server error, right after creating one.
	# Setting it here, once, before the main batch starts, closes
	# that race for the whole run (it also makes adminSettings.
	# feature's own snapshot see 'true' as the "current" value, so
	# its restore-at-the-end keeps it 'true' too, not 'false').
	- name: Seed platform settings
	env:
	BASE_URL: http://localhost
	run: yarn test:playwright:seed-settings

	# NOTE (2026-08-23): a fifth seed step ("Seed course
	# subscriptions", subscribing acostea/fapple to TEMP) briefly lived
	# here and was REMOVED again on purpose. It is redundant now that
	# playwright.config.ts pins `workers: 1`: file order is fixed, and
	# course_user_registration.feature sorts before toolGroup.feature,
	# so the subscriptions exist by the time anything needs them.
	#
	# It was also actively harmful. Hoisting those scenarios out of
	# their own file broke that file's internal narrative (it
	# subscribes amann, unsubscribes him, then re-subscribes him as a
	# fixture): with amann pre-subscribed, its first scenario had
	# nobody left to select, because subscribe_user.php's picker
	# excludes already-subscribed users. And it could never have fixed
	# the failures it targeted anyway — a seed guarantees a fixture
	# exists at the START, but course TEMP's user list is written by
	# four different feature files, so under parallel execution another
	# file could still unsubscribe fapple mid-run (measured: it did).
	# Ordering, not more seeding, is the actual fix. See
	# playwright.config.ts's `workers: 1` comment for the full history.

	# The step result is decided by check-results.mjs, NOT by
	# `playwright test`'s exit code. That is deliberate: the exit code is
	# non-zero for anything that went wrong anywhere in the process,
	# including work that is not a test, so it cannot distinguish "a test
	# regressed" from "a cleanup hook was slow". Twice in two days this
	# job went red on runs that were 410/410 green, because playwright-bdd
	# defers every tagged AfterAll into one worker-teardown phase that
	# blew its 90s budget AFTER the last test had already passed.
	#
	# A cleanup step running once every test has finished cannot
	# invalidate those tests, and two attempts at making that teardown fit
	# its budget both failed — so the durable fix is to stop asking the
	# exit code a question it cannot answer. The checker still fails the
	# build on any failed test, and on a TRUNCATED run (--min-tests),
	# which is the loophole a naive `\|\| true` would leave open. Non-test
	# errors are printed as warnings so they stay visible and fixable.
	- name: Playwright tests
	env:
	BASE_URL: http://localhost
	run: \|
	yarn test:playwright \|\| echo "::warning::playwright exited non-zero — deferring to the result check below"
	node tests/playwright/scripts/check-results.mjs \
	--report var/test-results/playwright/results.json \
	--min-tests 439 \
	--label "main batch"

	# MUST come before the SpecialCase1 step below, not at the end of
	# the job with the other uploads: `playwright test` wipes its own
	# outputDir at the START of every invocation, so the SpecialCase1
	# run would delete this batch's HTML report, traces and
	# screenshots before any end-of-job upload step could collect
	# them. Two separately-named artifacts is the only way to keep
	# both batches' evidence.
	- name: Upload main-batch Playwright report on failure
	if: failure()
	uses: actions/upload-artifact@v7
	with:
	name: playwright-report-main-php${{ matrix.php-versions }}
	path: \|
	var/test-results/playwright/report
	var/test-results/playwright/results
	var/test-results/playwright/results.json
	retention-days: 7
	if-no-files-found: ignore

	# SpecialCase1 runs in its own SEQUENTIAL step, excluded from the
	# parallel batch above via the @specialcase1 tag (see
	# package.json's --grep/--grep-invert pair). Three separate
	# reasons, any one of which would justify it:
	#
	# 1. ORDERING. specialCase1Sessions.feature hard-depends on
	# specialCase1PlatformSettings.feature having run first — the
	# session extra fields (extra_domaine/extra_theme_fr/
	# extra_theme_de/extra_ecouter) its 4 session scenarios fill in
	# only exist because "Add minimal session extra fields" creates
	# them, and its teacher-creation scenario needs
	# admins_can_set_users_pass, which that same file enables. In
	# the parallel batch there was NO ordering guarantee between
	# two different files at all. With --workers=1 and the config's
	# `fullyParallel: false`, files run in sorted order, and
	# "specialCase1PlatformSettings" sorts before
	# "specialCase1Sessions" — so the dependency is satisfied by
	# construction.
	# 2. COLLATERAL DAMAGE. specialCase1PlatformSettings mutates ~100
	# platform settings, several of which are global and UI-visible
	# on EVERY page (cookie_warning renders a fixed bottom banner
	# that intercepts pointer events; allow_registration/
	# allow_terms_conditions change the header on every page). Its
	# own "Tear down" scenario restores them, but only at the END —
	# so while it ran, every concurrently-executing file saw the
	# mutated platform. Running it alone closes that window
	# entirely.
	# 3. WORKER STARVATION. Its scenarios are tagged @long-scenario
	# (15-minute budget) and genuinely take minutes each. Dropping
	# several of those into the shared worker pool measurably
	# starved unrelated files: the CI run that first enabled them
	# produced fresh 90s timeouts in companyReports.feature and
	# accessCompanyReports.feature, which had passed immediately
	# before, plus a session-bleed failure in
	# admin/fileIntegrity.feature.
	#
	# `if: always()` so this still runs when the batch above fails —
	# SpecialCase1 derives from a real customer project, so its signal
	# must not be hidden behind an unrelated failure elsewhere. Placed
	# AFTER the batch (not before) so the batch sees pristine,
	# never-yet-mutated platform settings.
	# Same exit-code-vs-results reasoning as the main batch above.
	# --min-tests is this batch's own floor (11 scenarios), not the
	# suite-wide one — a shared floor would let a truncated SpecialCase1
	# run slip through.
	#
	# NOTE: test:playwright:specialcase1 deliberately passes
	# `--retries=0 --trace=retain-on-failure`, overriding the config's
	# `retries: 1` / `trace: "on-first-retry"`. Retries are WRONG for this
	# batch, and measured to be: a retry gets a fresh browser but the SAME
	# database, and these scenarios are heavily non-idempotent (they create
	# sessions, courses and extra fields). Observed on 2026-08-26 — "Add
	# minimal session extra fields" failed attempt 0 on the terms-and-
	# conditions interstitial, then attempt 1 failed with a COMPLETELY
	# different error ("Add courses to this session" not found) because
	# attempt 0 had already created the session. One clear failure became
	# two confusing ones.
	# `--trace=retain-on-failure` is the necessary partner: with retries
	# off, "on-first-retry" would never capture a trace at all, so a
	# SpecialCase1 failure would arrive with no evidence.
	# The main batch keeps retries — its scenarios largely clean up after
	# themselves, and there the retry is what absorbs the roving
	# `browserContext._wrapApiCall ... traceName` Playwright-internal error.
	- name: Playwright SpecialCase1 tests
	if: always()
	env:
	BASE_URL: http://localhost
	run: \|
	yarn test:playwright:specialcase1 \|\| echo "::warning::playwright exited non-zero — deferring to the result check below"
	node tests/playwright/scripts/check-results.mjs \
	--report var/test-results/playwright/results.json \
	--min-tests 11 \
	--label "SpecialCase1"

	# Whatever is left in outputDir at the end of the job — i.e. the
	# SpecialCase1 step's own output, since it runs last and wiped the
	# main batch's (uploaded separately above under
	# playwright-report-main-*). Kept on the original artifact name so
	# existing habits/tooling still find something here.
	- name: Upload Playwright report on failure
	if: failure()
	uses: actions/upload-artifact@v7
	with:
	name: playwright-report-php${{ matrix.php-versions }}
	path: \|
	var/test-results/playwright/report
	var/test-results/playwright/results
	var/test-results/playwright/results.json
	retention-days: 7
	if-no-files-found: ignore

	- name: Upload provisioning logs on failure
	if: failure()
	uses: actions/upload-artifact@v7
	with:
	name: provisioning-logs-php${{ matrix.php-versions }}
	path: /var/log/apache2/error.log
	retention-days: 7
	if-no-files-found: ignore

Navigation Menu

Update README to change chunk file suffix #108

Workflow file

Update README to change chunk file suffix #108

Uh oh!

Workflow file for this run