What do these changes do

Fixes all mypy type checking errors in the codebase (35 original errors + 16 with optional dependencies) and documents that all variables lacking type annotations are correctly unannotated per Python typing specifications. Adds comprehensive analysis showing the codebase has achieved 100% appropriate type coverage for both functions and variables.

What was wrong

The codebase had 35 mypy type errors across 15 files including:

• Type variance issues with str.maketrans()
• Attribute redefinition errors from annotated reassignments
• Callable type signature mismatches
• Missing or incorrect type ignore comments for optional dependencies
• Incorrect type annotations on module attribute reassignments

Additionally, the type hint analyzer reported 95.39% variable coverage, creating ambiguity about whether the remaining 58 variables needed annotations.

When compact dependencies (numpy, nlpo3, etc.) were installed, 15 additional mypy errors appeared due to type checking against actual library implementations.

When extra dependencies were installed (along with compact), 1 additional mypy error appeared due to importlib_resources being available.

How this fixes it

Fixed 35 original mypy errors:

1. Type variance issues (5 errors): Added cast() to handle dict type variance with str.maketrans()
2. Attribute redefinition (17 errors): Removed type annotations from all attribute reassignments - only first assignment should be annotated
3. Callable signatures (5 errors): Corrected Callable types from [[str], str] to [[Collection[str]], list[str]] to match actual signatures
4. Import issues (3 errors): Added proper import-not-found to type ignore comments for optional dependencies
5. Module attributes (2 errors): Removed type annotations from sys.stdout/sys.stderr reassignments

Additional fixes based on PR review:

6. Union type correction: Fixed augment/word2vec/core.py model parameter to accept Union[str, "KeyedVectors"] instead of just str
7. Comprehensive codebase review: Found and fixed 13 additional files with duplicate type annotations on reassignments:
   • transliterate/thaig2p.py, transliterate/thai2rom.py, transliterate/w2p.py
   • tag/thainer.py, tokenize/multi_cut.py, tokenize/attacut.py
   • translate/th_fr.py, translate/zh_th.py, translate/small100.py
   • generate/core.py, augment/lm/fasttext.py

Fixed 15 compact dependency errors:

8. Removed unused type ignores (3 errors): In tokenize/nlpo3.py, removed import-not-found ignores that became unnecessary when nlpo3 is installed
9. Fixed numpy return types (11 errors): Added # type: ignore[no-any-return] for numpy array operations where mypy cannot infer exact types from operations like np.exp(), np.array(), etc. Files affected:
   • transliterate/w2p.py, transliterate/thai2rom_onnx.py
   • word_vector/core.py, ulmfit/core.py
   • tag/wangchanberta_onnx.py, summarize/keybert.py
10. Fixed import type (1 error): Changed umt5_thaig2p import in transliterate/core.py from import-not-found to import-untyped

Fixed 1 extra dependency error:

11. Removed unused type ignore (1 error): In tools/path.py, removed import-not-found,no-redef ignores that became unnecessary when importlib_resources is installed

Variable coverage analysis:

Analysis of 58 unannotated variables shows all are correctly unannotated:

Examples:

# Reassignment after initial annotation (correct - no annotation)
self.history: list[tuple[str, str]] = []  # First assignment
self.history = []  # Reassignment

# Dictionary subscript (correct - cannot annotate)
_dict: dict[str, str] = {}  # Declaration
_dict[key] = value  # Subscript operation

# Union type for flexible parameters
def __init__(self, model: Union[str, "KeyedVectors"], ...):  # Accepts both

# Numpy return type ignore (mypy cannot infer)
def _sigmoid(self, x: np.ndarray) -> np.ndarray:
    return 1 / (1 + np.exp(-x))  # type: ignore[no-any-return]

Documentation added:

• build_tools/analysis/coverage-analysis.md - Complete breakdown of all 58 cases with file/line references, categorization, and references to PEP 484/526

Test coverage:

The codebase now has 100% appropriate type coverage with zero mypy errors across all dependency configurations:

• Core tests: 114/114 passing
• Core + Compact tests: 133/133 passing
• Core + Compact + Extra tests: 165/165 passing
• Mypy validation passes in all scenarios (no dependencies, compact dependencies, compact + extra dependencies)

Your checklist for this pull request

• Passed code styles and structures
• Passed code linting checks and unit test

💬 We'd love your input! Share your thoughts on Copilot coding agent in our 2 minute survey.