[jonathan343]

Summary

S3 uses CRC64NVME for full-object checksums, and it's the algorithm S3 computes when an upload doesn't specify one. This PR adds it to smithy-checksums as CRC64NVME, also available through new("crc64nvme"). It reuses the native extension and the pure-Python CRC implementation from #867. The native side uses crc-fast's Crc64Nvme. The pure-Python side reuses the table-driven CRC that CRC32C uses, with the CRC-64/NVME polynomial.

Example

import base64

import smithy_checksums

checksum = smithy_checksums.new("CRC64NVME")
checksum.update(b"Hello world")
print(checksum.name, checksum.digest_size, base64.b64encode(checksum.digest()))

The native and pure-Python builds print the same thing:

crc64nvme 8 b'OOJZ0D8xKts='

Testing

The known answers are the CRC-64/NVME check value for "123456789" and the four 64-bit CRC test cases from the NVMe NVM Command Set specification. Both implementations are also compared against the CRT and against each other:

uv run pytest packages/smithy-checksums -k "crc64nvme and (known_answers or matches_crt or native_matches_pure)" --no-cov -q

6 passed, 118 deselected in 0.11s

The chunked pure-Python CRC is also checked against a bit-at-a-time reference CRC across chunk boundaries:

uv run pytest packages/smithy-checksums -k "crc64nvme and chunked_crc_matches_bitwise" --no-cov -q

1 passed, 123 deselected in 0.14s

---

By submitting this pull request, I confirm that you can use, modify, copy, and redistribute this contribution, under the terms of your choice.

[stobrien89]