__setitem__() changes chunk dtype upon first access. This happens **before** the values are replaced, to avoid overflows or loss of de

Setting a bare string into a dataset doesn't swap the buffer dtype.

__setitem__ with an integer array index where the wholly-selected chunks are not contiguous among the selected chunks

Same bytes, different shape => different hash, in both implementations, and they agree with each other.

Test that shrinking a StagedChangesArray dereferences the *staged* slabs that are no longer referenced by any chunks. The full slab and the base s

Test that if a partial chunk is created or resized when shrinking, and that chunk was on a base slab, it is loaded into memory to avoid different

Test that a fancy index that can be redefined globally as a slice results in a SliceMapper

load() and changes() on a size-0 array that has base slabs

Test that versions can be staged and are the expected shape.

Test that stage_version logs stats after writing data.

Multiple chunks, with edge trimming and hash row permutation, carved out of a stepped (outer axis) slab: the chunk axis-0 offset must use the real

Test that writing an array of object strings without metadata correctly preserves metadata on the Dataset.

In h5py, it is OK to create and destroy astype("T") views in rapid succession, interleaved with __setitem__ calls of data with dtype="T":

Test that string chunk reuse works as intended.

Test that VersionedHDF5File.versions returns the same as all_versions(vfile).

Variable-width strings are hashed by their UTF-8 bytes, so ``str``, ``bytes`` and StringDType spellings of the same value collide *by design*.

Naive Python hashlib reimplementation of hash_slab.

Naive Python hashlib reimplementation of hash_slab for strings.