[FEATURE] Enable full repository-wide mypy type checking

[PredictiveManish]

What problem does this solve?

The LandmarkDiff repository currently has incomplete type annotations that prevent comprehensive mypy type checking. This creates several problems:

1. No type safety in scripts and benchmarks – The CI only runs mypy landmarkdiff/, leaving 125+ scripts and 3 benchmarks completely unchecked.

2. Errors are silently ignored – Most modules in pyproject.toml have ignore_errors = true, which means mypy runs but doesn't report actual type errors.

3. Pre-commit mypy hook is disabled – Due to duplicate module issues and type annotation gaps, mypy cannot run cleanly in pre-commit.

4. No real type checking happens – Despite mypy being configured, the combination of ignore_errors = true and limited scope means type errors go undetected.

Goal: Enable meaningful, comprehensive mypy type checking across the entire codebase to catch type-related bugs early and improve code quality.

Proposed solution

Phase 1: Configuration Updates

Update pyproject.toml mypy settings:

[tool.mypy]
python_version = "3.10"
warn_return_any = true
warn_unused_configs = true
disallow_untyped_defs = true
check_untyped_defs = true
explicit_package_bases = true
ignore_missing_imports = true
strict_optional = true
warn_redundant_casts = true
warn_unused_ignores = true

• Update CI to use parallel matrix for type checking:

  • landmarkdiff/ (main package)
  • scripts/ (all scripts)
  • benchmarks/ (benchmark scripts)

• Enable mypy in pre-commit with --explicit-package-bases flag to handle duplicate module names without file renames.

• Add package __init__.py files to directories that need them.

---

Phase 2: Remove ignore_errors = true (Module by Module)

The current pyproject.toml has ignore_errors = true for 33 modules. This will be removed incrementally:

• Start with simpler modules (e.g., api_client, config, log)
• Run mypy to identify type errors
• Add missing type annotations
• Verify module passes mypy
• Move to next module

Example fix:

# Before
def process_image(img, size):
    return cv2.resize(img, (size, size))

# After
def process_image(img: np.ndarray, size: int) -> np.ndarray:
    return cv2.resize(img, (size, size))

---

Phase 3: Scripts Type Annotations

After landmarkdiff is complete:

• Run mypy on scripts/ directory
• Add type annotations to all functions
• Fix any type errors

---

Phase 4: Benchmarks Type Annotations

Apply the same process to benchmarks/ directory.

Alternatives considered

No response

Area of the project

CLI / scripts

Would you be willing to implement this?

Yes, I'd like to submit a PR

References

#405 , #396

[PredictiveManish]

Will submit a PR for this, so anyone who's interested can find out other issues as it'll save your time.

[dreamlessx]

Great write-up @PredictiveManish! This is a solid proposal. A few notes:

1. Phased approach — start with landmarkdiff/ core package (remove ignore_errors = true incrementally), then extend to scripts/ and tests/. Trying everything at once will be overwhelming.

2. Priority modules — inference.py, landmarks.py, manipulation.py, and conditioning.py are the most critical for type safety since they form the core pipeline.

3. CI integration — once the core passes strict mypy, we can add it as a required CI check.

Looking forward to the PR! This complements your pre-commit hooks work nicely.

[PredictiveManish]

Great write-up @PredictiveManish! This is a solid proposal. A few notes:

1. Phased approach — start with landmarkdiff/ core package (remove ignore_errors = true incrementally), then extend to scripts/ and tests/. Trying everything at once will be overwhelming.
2. Priority modules — inference.py, landmarks.py, manipulation.py, and conditioning.py are the most critical for type safety since they form the core pipeline.
3. CI integration — once the core passes strict mypy, we can add it as a required CI check.

Looking forward to the PR! This complements your pre-commit hooks work nicely.

Yeah these steps is the same as I thought to work upon this, thanks for these priority modules will be more vigilant while working on this.

[dreamlessx]

Sounds good @PredictiveManish! Feel free to open a draft PR early so we can give feedback as you go. No rush — incremental is better than trying to fix everything at once.