Development manual

Keywords for commit messages

This section provides a list of keywords to include in commits comments. These keywords help organize and identify different types of contributions for better readability and maintainability.

Keywords List

<strong>1. <tt>doc:</tt>

• **Purpose: For documentation-related comments or detailed explanations.
• Example:

  // doc: General algorithm description for terrain contact

<strong>2. <tt>feat:</tt>

• **Purpose: To mark features or specific functionalities.
• Example:

  // feat: Added support for binary STL mesh files

<strong>3. <tt>fix:</tt>

• **Purpose: To identify bug fixes or adjustments.
• Example:

  // fix: Corrected input parameter validation

<strong>4. <tt>refactor:</tt>

• **Purpose: For restructuring code without changing its functionality.
• Example:

  // refactor: Separated normal calculation module

<strong>5. <tt>test:</tt>

• **Purpose: For comments related to tests or test cases.
• Example:

  // test: Validated force calculations in impact problem

<strong>6. <tt>perf:</tt>

• **Purpose: To highlight performance improvements or optimizations.
• Example:

  // perf: Reduced memory usage in explicit solver

<strong>7. <tt>todo:</tt>

• **Purpose: To indicate pending tasks or features.
• Example:

  // todo: Implement error handling for input keywords

<strong>8. <tt>hack:</tt>

• **Purpose: To identify temporary or workaround solutions.
• Example:

  // hack: Using fixed values to bypass null mass in the mesh nodes

<strong>9. <tt>warn:</tt>

• **Purpose: For warnings about potential issues or limitations.
• Example:

  // warn: Current update particle density method does not support particle shear changes

<strong>10. <tt>debug:</tt>

• **Purpose: For debugging or test-specific comments.
• Example:

  // debug: Prints the result of calculated normals

<strong>11. <tt>chore:</tt>

• **Purpose: For minor maintenance or cleaning tasks.
• Example:

  // chore: Removed unused code in mesh reading module

<strong>12. <tt>deprecated:</tt>

• **Purpose: To mark features or methods that are no longer in use.
• Example:

  // deprecated: This method will be removed in the next version

Best Practices

1. Consistency:
   • Always use these keywords in lowercase to maintain clarity.
2. Combination:
   • Combine keywords if necessary to clarify intent. For example:

     // feat: Added error handling for invalid inputs
     // fix: Corrected edge case with empty arrays

Parallelization Guidelines

Parallelization Guidelines in MPM-Geomechanics

This project uses OpenMP to accelerate selected functions through multi-threaded parallelization.

To maintain clarity and avoid unnecessary code duplication, we adopt the following convention:

---

Use macro-controlled dual versions <strong>only when needed</strong>

Separate sequential and parallel versions (using #if defined(...)) should be used only when the function requires thread-local storage such as:

• Mass or force accumulation into nodal values
• Boolean flags (e.g. node activation)
• Temporary vectors shared across threads

Examples:

• nodalMass
  
• nodalMomentum
  
• nodalInternalForce

---

Use a single-version pattern with conditional pragma otherwise

If the function does not require thread-local storage, and the parallel and sequential logic is the same, then use this structure:

#ifdef _OPENMP
#pragma omp parallel for
#endif
for (int i = 0; i < N; ++i) {
// loop body
}

Use of <tt>static_cast<int></tt> in <tt>for</tt> loops

In several parts of the simulator, loops iterate over particle or node containers whose size is returned as size_t. Since OpenMP requires a signed integer loop counter, we must convert size_t to int.

To ensure clarity and safety, always use static_cast<int> instead of the C-style (int) cast.

Rationale:

• static_cast<int> is the C++ standard and makes the conversion intention explicit.
  
• It prevents unsafe or ambiguous conversions allowed by the C-style cast.
  
• It improves code readability and consistency across the project.

Example (correct usage):

#ifdef _OPENMP
#pragma omp parallel for
#endif
for (int i = 0; i < static_cast<int>(particles->size()); ++i) {
    if (!particles->at(i)->getActive()) continue;
    // ... //
}

Normalization of OpenMP pragmas

All parallel loops must be guarded with #ifdef _OPENMP to ensure clean compilation with or without OpenMP support. Always use the same style:

#ifdef _OPENMP
#pragma omp parallel for schedule(static)
#endif
for (int i = 0; i < static_cast<int>(particles->size()); ++i) {
    // loop body
}

Atomic operations: always take the reference first

When using #pragma omp atomic, always assign the target variable to a reference before the atomic operation:

double& m = nodeI->getMassRef();
#ifdef _OPENMP
  #pragma omp atomic update
#endif
m += nodalMass;