D657 - Both warnings were mirrors

Status: measured Date: 2026-09-09

The contradiction was inside this crate

REQ-20260909T1910Z-3d05 asked SELFish which of the two projects' documents was describing the other's convention. They read this crate's source and answered: neither against the other's - it is orbistoun's against orbistoun's, one function apart.

NidHasher::hash        "read little-endian"
NidHasher::hash_bytes  "Big-endian, and this was wrong for a long time"
                       u64::from_be_bytes(...)

The implementation is hash_bytes and always was. The wording on hash was simply wrong, and had been sitting one screen above the paragraph that contradicted it.

A test vector would not have found this. 3d05 offered one, and it would have confirmed what both projects already agreed on - that the bytes match once reversed - while leaving a wrong sentence in place. What found it was somebody reading the source rather than taking the question at face value.

The naming, which is the part worth keeping

Both docs now say it without an endian word:

The first eight bytes of the SHA-1 digest, packed so that digest[0] is the most significant byte.

"Little-endian" and "big-endian" describe the packing, and the packing is exactly what differs between the two projects. So each reached for the word that described its own, and called the other one incorrect - in mirror-image notes, in two repositories, about the same eight bytes. SELFish's data file said orbistoun's big-endian claim "is incorrect"; orbistoun's hash_bytes said reading it the other way "agrees with nothing". Both were true locally and both were phrased as though universal.

digest[0] is the most significant byte says which end without reference to whose machine or whose u64, and cannot be mirrored.

What was actually wrong, and what was not

Neither convention is a bug. The u64 is an internal representation; each project unpacks the same digest its own way and arrives at the same eleven characters. What matters is the pairing

that a crate hashes and decodes the same way round - and encode_nid exists mainly so a round-trip test pins exactly that (D070).

So hash_bytes now says "wrong here", with the reason the qualifier is not modesty. A warning written as universal is a warning that will be contradicted by somebody equally careful, and then one of the two gets believed for the wrong reason.

The shape this keeps taking

Three times in two days a confident report turned out to be about units rather than about substance: the 0x6bc000 table addresses, the four harness errors, and now this. Every one was two things that shared a name and not a meaning, and every one produced a report that named a defect somewhere. The differential's own comments carry the first two; this crate's now carries the third.