pFUnit Troubleshooting

This skill provides comprehensive guidance for debugging and fixing pFUnit test failures in MAPL. pFUnit is the parallel Fortran unit testing framework used throughout MAPL.

Quick Reference

Common pFUnit Commands

# Run single test with diagnostics
./TestName.x -d

# Run filtered tests (substring match)
./TestName.x -f substring

# Run with verbose output
./TestName.x -v

# List all tests
./TestName.x -l

# Run specific test by exact name
./TestName.x -f exact_test_name

Fast Generic3g Test Workflow

# From MAPL build directory (e.g., ~/swdev/VS/MAPL/nag)
cd generic3g/MAPL_cfio/MAPL_cfio_r4.tests

# Set library path (required for NAG)
export DYLD_LIBRARY_PATH=/usr/local/nag/lib:$DYLD_LIBRARY_PATH

# Run tests (< 1 minute)
./generic3g.x

Understanding pFUnit Test Structure

Test Executable Naming

MAPL uses consistent naming for test executables:

• •generic3g.x - Generic 3D grid tests
• •TestName.x - Other component tests

Test Organization

Tests are organized by component:

build-dir/
├── base/
│   └── tests/
│       └── TestSomething.x
├── generic3g/
│   └── MAPL_cfio/
│       └── MAPL_cfio_r4.tests/
│           └── generic3g.x
├── gridcomps/
│   └── tests/
└── ...

pFUnit Test File Extensions

• •*.pf - pFUnit test source files (preprocessed to .F90)
• •*.F90 - Generated or manual Fortran test files
• •*.x - Compiled test executables

Debugging Test Failures

Step 1: Identify the Failure

When ctest reports a failure:

# Run ctest to see which tests fail
ctest

# Example output:
# 99% tests passed, 1 tests failed out of 100
# The following tests FAILED:
#     42 - generic3g (Failed)

Step 2: Run Test Directly with Diagnostics

Navigate to test directory and run with diagnostics:

# Find the test executable
find . -name "generic3g.x" -type f

# Navigate to test directory
cd generic3g/MAPL_cfio/MAPL_cfio_r4.tests

# Run with diagnostics
./generic3g.x -d

The -d flag shows:

• •Test names as they execute
• •Assertion failures with line numbers
• •Expected vs actual values
• •Stack traces (if available)

Step 3: Run Specific Failing Test

Use -f to filter to specific test:

# If test is named "test_vertical_interpolation"
./generic3g.x -f vertical_interpolation

# Multiple matches possible (substring matching)
./generic3g.x -f interpolation

IMPORTANT: The -f flag does substring matching, NOT wildcard/glob matching. Do not use * or ?.

Step 4: Examine Test Output

Common failure patterns:

Assertion Failures

Location: test_vertical_interpolation.pf:45
 Expected: 1.23456
 Actual:   1.23457
 Difference exceeds tolerance

What to check:

• •Is the tolerance too strict?
• •Is there a numerical precision issue?
• •Did algorithm change affect expected values?

Segmentation Faults

Program received signal SIGSEGV: Segmentation fault

What to check:

• •Array bounds violations
• •Uninitialized pointers/allocatables
• •Memory corruption

Hangs/Timeouts

Test runs but never completes.

What to check:

• •Infinite loops
• •Deadlocks in parallel code
• •MPI issues

Common pFUnit Issues and Solutions

Issue: "No such test"

Symptom:

./generic3g.x -f my_test
# No tests run

Solution: List all available tests:

./generic3g.x -l

Check exact test name and use correct substring.

Issue: Library Loading Errors (macOS)

Symptom:

dyld: Library not loaded: @rpath/libnag_nag.dylib

Solution (NAG on macOS):

export DYLD_LIBRARY_PATH=/usr/local/nag/lib:$DYLD_LIBRARY_PATH
./generic3g.x

Solution (gfortran on macOS):

export DYLD_LIBRARY_PATH=/usr/local/gfortran/lib:$DYLD_LIBRARY_PATH
./generic3g.x

IMPORTANT: This is required for direct test execution. Not needed for ctest if properly configured.

Issue: Test Passes Locally, Fails in CI

Possible causes:

1. •Compiler differences - CI may use different compiler
2. •Timing issues - CI may be slower/faster
3. •Environment differences - Different library versions
4. •Uninitialized variables - Different compilers handle differently

Debugging approach:

1. •Check which compiler CI uses
2. •Build locally with same compiler
3. •Run tests with that compiler
4. •Compare results

Issue: Numerical Tolerance Failures

Symptom:

Expected: 1.2345678
Actual:   1.2345679

Common causes:

• •Compiler optimization differences
• •Floating-point rounding in different orders
• •Platform differences (x86 vs ARM)

Solutions:

1. •

   Relax tolerance (if difference is negligible):

   fortran

   @assertEqual(expected, actual, tolerance=1.0e-5)
   

2. •

   Use relative tolerance:

   fortran

   @assertEqual(expected, actual, tolerance=abs(expected)*1.0e-6)
   

3. •

   Investigate algorithm if difference is significant

Issue: MPI-Related Failures

Symptom: Test fails only when run with multiple processes.

Debugging approach:

1. •

   Run with single process:

   bash

   mpirun -np 1 ./TestName.x
   

2. •

   Run with multiple processes:

   bash

   mpirun -np 4 ./TestName.x
   

3. •

   Check for:

   • •Race conditions
   • •Incorrect MPI communication patterns
   • •Uninitialized data on some ranks

Issue: Test Executable Not Found

Symptom:

./TestName.x
bash: ./TestName.x: No such file or directory

Solution: Test may not have been built. Rebuild:

cd <build-directory>
make -j8 install

If still not found, test may be disabled or have build errors. Check CMake output.

Issue: Tests Fail After Code Changes

Debugging workflow:

1. •

   Verify change didn't break assumptions:

   • •Did you change function signatures?
   • •Did you modify behavior that tests depend on?
2. •

   Update tests if behavior intentionally changed:

   fortran

   ! Update expected values
   @assertEqual(new_expected_value, actual)
   

3. •

   Fix code if tests revealed bugs:

   • •Tests are correct, code is wrong
   • •This is the most common case!
4. •

   Check for stale build artifacts:

   bash

   make clean
   make -j8 install
   ctest
   

Writing Effective pFUnit Tests

Basic Test Structure

@test
subroutine test_my_function()
   use pfunit_mod
   use MyModule
   implicit none
   
   integer :: result
   integer :: expected
   
   ! Setup
   expected = 42
   
   ! Execute
   result = my_function(input)
   
   ! Verify
   @assertEqual(expected, result)
   
end subroutine test_my_function

Common Assertions

! Equality assertions
@assertEqual(expected, actual)
@assertEqual(expected, actual, tolerance=1.0e-6)
@assertEqual(expected, actual, message="Custom error message")

! Boolean assertions
@assertTrue(condition)
@assertFalse(condition)

! Exception/error assertions
@assertExceptionRaised("Expected error message")

! Array assertions
@assertEqual(expected_array, actual_array)
@assertEqual(expected_array, actual_array, tolerance=1.0e-6)

Setup and Teardown

@before
subroutine setUp()
   ! Initialize test fixtures
   ! Allocate arrays
   ! Open files
end subroutine setUp

@after
subroutine tearDown()
   ! Clean up test fixtures
   ! Deallocate arrays
   ! Close files
end subroutine tearDown

@test
subroutine test_something()
   ! setUp() called automatically before this
   ! Test code here
   ! tearDown() called automatically after this
end subroutine test_something

Testing with Floating-Point Numbers

CRITICAL: Never use exact equality for floating-point:

! BAD - fragile, compiler-dependent
@assertEqual(1.23456, result)

! GOOD - uses tolerance
@assertEqual(1.23456, result, tolerance=1.0e-5)

! BETTER - relative tolerance for large numbers
@assertEqual(expected, result, tolerance=abs(expected)*1.0e-6)

Testing Arrays

@test
subroutine test_array_operation()
   use pfunit_mod
   implicit none
   
   real, allocatable :: expected(:)
   real, allocatable :: result(:)
   integer :: i
   
   allocate(expected(10))
   allocate(result(10))
   
   ! Setup expected values
   do i = 1, 10
      expected(i) = real(i) * 2.0
   end do
   
   ! Call function under test
   call compute_array(result)
   
   ! Verify
   @assertEqual(expected, result, tolerance=1.0e-6)
   
   deallocate(expected)
   deallocate(result)
   
end subroutine test_array_operation

Debugging Techniques

Using Print Statements

Quick debugging in pFUnit tests:

@test
subroutine test_with_debug_output()
   use pfunit_mod
   implicit none
   
   integer :: result
   
   result = my_function(42)
   
   print *, "DEBUG: result =", result
   print *, "DEBUG: expected =", 84
   
   @assertEqual(84, result)
   
end subroutine test_with_debug_output

Run test to see debug output:

./TestName.x -d -f test_with_debug_output

Using GDB with pFUnit

# Compile with debug symbols (should already be in Debug build)
cmake .. -DCMAKE_BUILD_TYPE=Debug
make -j8 install

# Run test under GDB
gdb ./TestName.x

# In GDB:
(gdb) run -f failing_test
(gdb) bt  # backtrace when it crashes
(gdb) print variable_name

Using LLDB (macOS)

lldb ./TestName.x

# In LLDB:
(lldb) run -f failing_test
(lldb) bt  # backtrace
(lldb) frame variable  # show local variables

Compiler Debug Flags

Different compilers offer different debugging capabilities:

NAG:

# Already has excellent runtime checking in Debug mode
cmake .. -DCMAKE_BUILD_TYPE=Debug

gfortran:

# Add extra runtime checks
cmake .. -DCMAKE_BUILD_TYPE=Debug \
  -DCMAKE_Fortran_FLAGS="-g -fbacktrace -fcheck=all"

Intel:

# Add runtime checks
cmake .. -DCMAKE_BUILD_TYPE=Debug \
  -DCMAKE_Fortran_FLAGS="-g -traceback -check all"

Performance Testing

Measuring Test Execution Time

# Time entire test suite
time ./TestName.x

# Time specific test
time ./TestName.x -f specific_test

Identifying Slow Tests

# Run with verbose output to see timing
ctest -V

# Or use ctest's built-in timing
ctest --output-on-failure --verbose

If a test is unexpectedly slow:

1. •Check for I/O operations (file reads/writes)
2. •Check for large array allocations
3. •Check for nested loops with high iteration counts
4. •Consider if test is actually a performance test (intentionally slow)

Test Maintenance Best Practices

When to Update Tests

1. •Function behavior changed intentionally → Update test expectations
2. •Bug was found in test → Fix the test
3. •Implementation optimized but behavior unchanged → Tests should still pass
4. •Test is flaky → Fix test to be deterministic

When to Disable Tests

CRITICAL: Disabling tests should be rare and temporary.

Valid reasons:

• •Test depends on external resource that's unavailable
• •Test is known to be flaky (GitHub issue filed)
• •Feature being tested is deprecated

How to disable:

@test(ifdef=SOME_FEATURE_FLAG)
subroutine test_to_disable()
   ! This test only runs if SOME_FEATURE_FLAG is defined
end subroutine test_to_disable

Or remove from CMakeLists.txt temporarily (prefer ifdef approach).

Test Naming Conventions

Good test names are descriptive:

! GOOD
@test
subroutine test_vertical_interpolation_linear_profile()

@test  
subroutine test_grid_create_with_invalid_dimensions_should_fail()

! BAD (too vague)
@test
subroutine test1()

@test
subroutine test_grid()

Integration with GitHub Workflow

Before Creating PR

# Run full test suite locally
cd <build-directory>
ctest

# If any failures, debug and fix
cd path/to/failing/test
./TestName.x -d

# Fix code or test, rebuild, retest
make -j8 install
ctest

CRITICAL: All tests must pass before creating PR.

Responding to CI Test Failures

If tests pass locally but fail in CI:

1. •Check CI logs for exact failure
2. •Note compiler/platform used in CI
3. •Reproduce locally with same compiler
4. •Fix and push updated code

See github-workflow skill for PR process.

Related Skills

• •mapl-testing - General MAPL testing workflows and strategies
• •mapl-build - Building MAPL (required before testing)
• •fortran-style - Fortran coding standards (affects test code too)
• •mapl-error-handling - Error handling in code being tested
• •github-workflow - CI/CD and PR process

Advanced Topics

Parallel Test Debugging

When debugging MPI-parallel tests:

# Run under MPI with debugger
mpirun -np 4 xterm -e gdb ./TestName.x

This opens separate debugger windows for each MPI rank.

Memory Debugging with Valgrind

# Check for memory leaks
valgrind --leak-check=full ./TestName.x -f specific_test

# Check for uninitialized memory
valgrind --track-origins=yes ./TestName.x -f specific_test

Note: Valgrind may report false positives with MPI and OpenMP.

Testing Private Module Procedures

pFUnit tests can't directly test private procedures. Options:

1. •Test through public interface (preferred)
2. •Make procedure public for testing (acceptable if documented)
3. •Use test-only wrapper (complex, rarely needed)

Parameterized Tests

For testing multiple similar cases:

@test(cases=[1, 2, 3, 4, 5])
subroutine test_with_parameter(n)
   integer, intent(in) :: n
   
   ! Test using parameter n
   @assertEqual(n * 2, my_double_function(n))
   
end subroutine test_with_parameter

When to use this skill:

• •Debugging test failures
• •Understanding pFUnit output
• •Writing new tests
• •Fixing flaky tests
• •Responding to CI failures