Coverage Metrics Strategy

When to Use

Measuring how much of your code is executed during tests, identifying untested code paths, setting coverage targets for CI/CD.

Decision: PCOV vs Xdebug

Criteria	PCOV	Xdebug
Speed	2-5x faster (15s vs 50s for typical suite)	Slower but acceptable for thorough analysis
Coverage types	Line coverage ONLY	Line, branch, path, function coverage
Setup complexity	Simple (single PHP extension)	Moderate (configure modes)
Maintenance status	Not actively maintained (last release 2021)	Actively developed, well-supported
PHP 8.4+ compatibility	Requires patches	Native support
Use case	Daily dev workflow, CI line coverage	Deep analysis, branch/path coverage

Recommendation: Use PCOV for fast feedback in development and CI/CD (line coverage is 90% of value). Use Xdebug when branch/path coverage matters (security-critical code, complex conditionals). For projects targeting PHP 8.4+, prefer Xdebug due to PCOV maintenance concerns.

Coverage Types Explained

Type	What it measures	Example	Why it matters
Line coverage	% of executable lines run	50 lines total, 40 executed = 80%	Quick indicator of test breadth
Branch coverage	% of conditional branches taken	`if ($x) { A } else { B }` -- both A and B executed?	Finds untested edge cases
Path coverage	% of unique execution paths	Function with 3 ifs = 8 possible paths	Ensures all combinations tested
Function coverage	% of functions/methods called	10 methods, 8 called = 80%	Identifies dead code

Line coverage is the baseline -- fast to collect, easy to understand. Branch coverage catches the "happy path only" problem. Path coverage is expensive but critical for security-sensitive code. Function coverage is a sanity check (0% = completely untested class).

Pattern

Installing PCOV (DDEV example):

# In .ddev/config.yaml
webimage_extra_packages: [php-pcov]

# Or manually
ddev exec sudo apt-get install php-pcov
ddev exec sudo phpenmod pcov
ddev restart

Installing Xdebug (DDEV includes it by default):

ddev xdebug on  # Enable Xdebug in coverage mode

Running PHPUnit with line coverage (PCOV or Xdebug):

# HTML report (open reports/index.html in browser)
./vendor/bin/phpunit --coverage-html reports/

# Clover XML (for CI/CD tools like GitLab, SonarQube)
./vendor/bin/phpunit --coverage-clover coverage.xml

# Text report (terminal output)
./vendor/bin/phpunit --coverage-text

# Filter to specific directory
./vendor/bin/phpunit --coverage-html reports/ web/modules/custom/my_module/

Running PHPUnit with branch/path coverage (Xdebug only):

# Requires Xdebug 3.1+
XDEBUG_MODE=coverage ./vendor/bin/phpunit \
  --coverage-html reports/ \
  --path-coverage

Interpreting coverage reports:

Code Coverage Report:
  Classes: 85.00% (17/20)
  Methods: 78.57% (22/28)
  Lines:   82.35% (140/170)

MyService.php
  Lines: 92.31% (12/13) -- GOOD
  Untested line 45: Exception handler

DiscountCalculator.php
  Lines: 50.00% (5/10) -- NEEDS WORK
  Untested lines 20-24: Edge case validation

Coverage in CI/CD (GitLab CI example):

test:
  script:
    - composer install
    - ./vendor/bin/phpunit --coverage-text --colors=never
  coverage: '/^\s*Lines:\s*\d+.\d+\%/'
  artifacts:
    reports:
      coverage_report:
        coverage_format: cobertura
        path: coverage.xml

Coverage Targets by Code Type

Code type	Target coverage	Rationale
Services (business logic)	80-90%	High reuse, critical to functionality
Plugins (blocks, formatters)	70-85%	Well-defined contracts, moderate complexity
Controllers	60-75%	HTTP overhead, test logic not routing
Forms (validation/submit)	50-70%	Setup expensive, focus on custom logic
Hooks/event subscribers	60-80%	Integration points, side effects matter
Entity classes (custom)	70-85%	Data integrity critical
Utility classes	90%+	Pure functions, easy to test thoroughly
Admin UI / config forms	30-50%	Mostly CRUD, low custom logic

Why 100% is a waste: Chasing 100% coverage leads to testing trivial getters/setters, framework code, and defensive branches that never execute. Meaningful coverage tests behavior, not code structure.

Common Coverage Anti-Patterns

Testing getters/setters -- 100% line coverage, 0% value (trivial code)
Covering dead branches -- if (FALSE) { unreachable } counts toward coverage but tests nothing
Ignoring branch coverage -- 80% line coverage but only "happy path" tested, failures uncaught
Coverage theater -- hitting lines without asserting behavior (test runs code but verifies nothing)
Excluding too much -- --coverage-exclude hides real gaps instead of writing tests
Coverage targets without context -- 80% of what? If 80% excludes all security checks, target is meaningless