The np.ascontiguousarray wrapper is only applied to the MFI function, but not to other similar volume-based indicators like AD, OBV, and ADOSC in src/numta/api/volume_indicators.py, which also use numba JIT compilation and accept the same types of inputs.

If non-contiguous arrays cause segfaults in MFI's numba functions, the same issue could potentially affect other indicators. Consider either:

1. Applying this fix consistently to all similar numba-compiled functions that accept multiple array inputs, or
2. Documenting why MFI specifically requires contiguous arrays while others don't

This ensures consistency and prevents similar issues from occurring with other indicators.

The double wrapping np.ascontiguousarray(np.asarray(...)) is inefficient. Consider using np.ascontiguousarray directly with the dtype parameter instead:

high = np.ascontiguousarray(high, dtype=np.float64)
low = np.ascontiguousarray(low, dtype=np.float64)
close = np.ascontiguousarray(close, dtype=np.float64)
volume = np.ascontiguousarray(volume, dtype=np.float64)

This avoids creating an intermediate array when the input is not already a numpy array, improving performance while achieving the same result.

The explicit dtype casting here is redundant. The pandas accessor's _get_column method (line 119 in src/numta/pandas_ext.py) already converts DataFrame column values to np.float64 using .values.astype(np.float64). Additionally, the MFI function itself now ensures correct dtype conversion with np.asarray(..., dtype=np.float64).

This redundant casting adds unnecessary overhead and makes the test inconsistent with other volume indicator tests (test_obv_calculation and test_ad_calculation) which don't perform explicit casting.