[RaphaelFakhri]

Fletcher32.encode raises IndexError for an empty buffer, and Fletcher32.decode raises IndexError for a buffer that holds only the 4-byte checksum. The codec takes &mv[0] of an empty memoryview, which fails Cython's bounds check.

This change:

• Returns a checksum of 0 from _fletcher32 for empty input, which matches the HDF5/netCDF reference implementation.
• Skips the payload memcpy in encode and in the out branch of decode when the payload is empty.
• Adds test_empty, which round-trips an empty buffer with and without out.

Closes #888

• Unit tests and/or doctests in docstrings
• Tests pass locally
• Docstrings and API docs for any new/modified user-facing classes and functions
• Changes documented in docs/release.md
• Docs build locally
• GitHub Actions CI passes
• Test coverage to 100% (Codecov passes)

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 89.71%. Comparing base (4afa628) to head (db932db).

Additional details and impacted files

@@           Coverage Diff           @@
##             main     #889   +/-   ##
=======================================
  Coverage   89.71%   89.71%           
=======================================
  Files          27       27           
  Lines         914      914           
=======================================
  Hits          820      820           
  Misses         94       94           

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[Duo-ee]

The empty-buffer fix looks right, but decode still has no size check on out, and that one is a memory-safety problem rather than a silent repeat:

• out smaller than b_len - FOOTER_LENGTH → memcpy writes past the destination (heap corruption / crash).
• out larger → the payload is copied, the tail of the caller's buffer is left untouched, and decode still reports success.
• ensure_contiguous_ndarray(out) is np.asarray(out) for non-array buffers, so a bytearray out becomes read-only and &out_mv[0] points at read-only memory.

A ValueError before the copy covers all three cases:

payload_len = b_len - FOOTER_LENGTH
out_mv = ensure_contiguous_ndarray(out).view("uint8")
if out_mv.size != payload_len:
    raise ValueError(
        f"Output buffer has {out_mv.size} bytes, but the decoded payload has {payload_len} bytes."
    )
memcpy(&out_mv[0], b_ptr, payload_len)

Worth adding a case with an oversized out and a short one next to test_empty.

[RaphaelFakhri]

@Duo-ee The undersized out overflow is real, but it's on main already and has nothing to do with empty buffers, so I'd rather keep this PR to #888. It fits with your jenkins out size PR, and a matching fix for Fletcher32 (and any other codec that copies into out by hand) can go in as its own PR.

Two notes on the details. A bytearray out is fine: np.asarray(bytearray(...)) is writable, and out_mv is a non-const typed memoryview, so a read-only buffer raises ValueError on assignment instead of being written to. And the snippet above takes &out_mv[0] unconditionally, which brings back the IndexError for an empty payload that this PR fixes.