Say what a vector quantity's bytes are

[matt-edmondson]

The reflection table's representation says what the bytes of a member are, and for a quantity it said the number the quantity is stored in. For a magnitude that is right — a Mass over a float is a float. For a vector form it is false: a Velocity3D over a float is three floats, which is a Vector3.

False in the way that reads as true, which is what makes it a fix rather than a note.

How it goes wrong

A consumer walking the table asks the representation how many lanes a member has. Holotype's validator does exactly that:

case TypeKind::Vector2:
case TypeKind::Vector3:
case TypeKind::Vector4: {
    const std::uint8_t count = detail::lanes<TypeKind>(member.representation);
    const std::uint32_t lane_size = member.size / count;

Told Float, a Velocity3D never reaches that arm at all — it takes the scalar one, which reads the member's first four bytes as the whole value and checks that against the range while the other two components go unexamined. Its network codec's "is this numeric" test has the same shape. A check that quietly examines a third of what it was given is worse than one that declines.

I found this pinning Holotype's reflection_test.cpp, which asserts kind and representation for each member — the assertions this would have silently changed the meaning of.

The fix

A quantity of two, three or four components reports Vector2, Vector3 or Vector4; one of zero or one reports its storage. The arity comes from QuantityRegistry, which reads it off the IVectorN the quantity implements, so it cannot disagree with the type it describes.

kind is unaffected — all of them are still Quantity. That is the whole point of the table carrying both: kind is what the schema said, representation is what the bytes are.

Tests

QuantityCppTests gains TheTableSaysWhatAQuantitysBytesAre, which generates a class holding one quantity of each arity and asserts that all of them are kind = Quantity while the representations differ — so the distinction is pinned rather than implied.

Every arity is covered rather than a sample of them, and every one is a real quantity rather than defensive code: the vocabulary has 148 magnitudes, 27 signed scalars, and 8, 22 and 7 quantities of two, three and four components, so Mass, Heading, Velocity2D, Velocity3D and Velocity4D reach all four arms of the mapping.

That second commit is what took coverage on new code from 75% to 100%, which is what SonarCloud's gate first failed on — but the arms were worth covering on their own account, since the rule the table states is only half-stated by asserting two of its four cases. It is verified by regression rather than by the gate: with the Vector2 arm returning the storage instead, Drift reports Float and the test fails.

Schema.Cpp.Test is 102 of 102 locally, and CI is green on all three platforms.

Context

This is the third in the chain behind matt-edmondson/Holotype#16, which merged and left Holotype's main red. Holotype needs a kschema with this in it before its reflection_test can assert anything true, so I would rather it pin one version than pin 1.34.1 and then have to change the assertions again.

🤖 Generated with Claude Code

https://claude.ai/code/session_01UGHDsYaaTQdzVR4XBR6miu

---

Generated by Claude Code

[sonarqubecloud]

Quality Gate passed

Issues
0 New issues
0 Accepted issues

Measures
0 Security Hotspots
100.0% Coverage on New Code
0.0% Duplication on New Code

See analysis details on SonarQube Cloud