feat(v3): float4/float8 encrypted-domain types

[tobyhede]

Summary

Adds IEEE-754 binary-float encrypted-domain coverage (real/float4 and double precision/float8) to the eql_v3 scalar surface, fully wired through codegen, the SQLx matrix oracle, and an edge-case regression suite. Implements the committed plan (docs/plans/2026-06-19-float-encrypted-domain-plan.md).

Both widths are ordered scalars reusing the existing four-domain shape, and both encrypt through a single f64 crypto path (Plaintext::Float(Some(x as f64))) — float4 vs float8 is purely a Postgres-surface distinction; the ciphertext/ORE term are byte-identical. Float is a new ScalarKind (F32/F64); the kind-agnostic codegen renderers generate the SQL with no codegen change (the flagged primary risk — confirmed a false alarm).

What's in here (one commit per task)

• Catalog — ScalarKind::F32/F64, Fixture::Float, FLOAT4/FLOAT8 specs + invariant tests.
• Harness — F4(f32)/F8(f64) newtypes (Ord via total_cmp, #[sqlx(transparent)]), value accessors via the shared lazy_values! macro, hand-written ScalarType/OrderedScalar/SignedScalar, guards (incl. an f32→f64 widening-monotonicity guard).
• Crypto path — EqlPlaintext for F4/F8 → Plaintext::Float; Cast::REAL/DOUBLE, PlaintextSqlType::REAL/DOUBLE_PRECISION.
• Routing — proc-macro is_float_token + float fixture discriminator, scalar_fixture! float arm, matrix dispatch lines.
• Tests — fixture-oracle + e2e-oracle suites, sign-boundary tests, a cross-width byte-identity test (F4(x) and F8(x as f64) share hm/ob), and a float_special suite (NaN / ±0 / ±Inf) that encrypts fresh so NaN never enters the shared fixture table.
• Artefacts/docs — committed tests/codegen/reference/{float4,float8}/ baselines, CHANGELOG entry, docs/upgrading/v2.4.md (U-001).

Verified locally (no DB required)

• cargo test -p eql-scalars (71), cargo test -p eql-tests-macros (22, incl. 2 float tests)
• cargo test -p eql_tests --lib float guards + EqlPlaintext float tests (11) + fixture-module spec tests
• mise run test:matrix:inventory — float4 → ordered, float8 → ordered, catalog reconciled
• mise run codegen:parity, mise run test:codegen (71), mise run clean && mise run build, mise run test:self_contained_v3
• Full encrypted_domain binary compiles under --features proptest-e2e
• Fixture generation through real cipherstash encryption (eql_v2_float4.sql/eql_v2_float8.sql, 13 rows each)

Deferred to CI (needs Postgres + ZeroKMS, unavailable locally)

The DB-backed runs — fixture/e2e oracles, sign-boundary, float_special, and the full matrix — run in CI. (Local Docker was unusable this session.)

Note: included one pre-existing fix

1f605a1f syncs matrix_tests_storage_only.txt with the storage_count_distinct_extractor arm that tests/sqlx/src/matrix.rs already emits — pre-existing drift on eql_v3 that blocked test:matrix:inventory on bool (unrelated to float). Regenerated via the documented command.

Summary by CodeRabbit

• New Features
  • Added support for encrypted floating-point values via new eql_v3.float4 and eql_v3.float8 ordered encrypted-domain types.
  • Enabled equality and full comparison operations (<, <=, >, >=) on encrypted float columns.
  • Provided TypeScript type bindings and JSON schemas for float encrypted domains and their comparison variants.

[coderabbitai]

Important

Review skipped

Auto reviews are disabled on base/target branches other than the default branch.

Please check the settings in the CodeRabbit UI or the .coderabbit.yaml file in this repository. To trigger a single review, invoke the @coderabbitai review command.

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Pro

Run ID: f703fb56-c0ff-4f9b-8f5a-91d1cd0e1e1f

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

• ✅ Review completed - (🔄 Check again to review again)

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch eql_v3_float_domains

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands and usage tips.}

[coderabbitai]

Actionable comments posted: 1

🧹 Nitpick comments (1)

tests/codegen/reference/float4/float4_ord_ore_operators.sql (1)

118-146: Add COMMUTATOR declarations to containment operators.

The @> and <@ operators are logical commutators—they express reciprocal containment relationships—but these operator overloads omit the COMMUTATOR metadata. This prevents the planner from inferring symmetry optimizations.

Suggested fix

 CREATE OPERATOR @> (
   FUNCTION = eql_v3.contains,
-  LEFTARG = eql_v3.float4_ord_ore, RIGHTARG = eql_v3.float4_ord_ore
+  LEFTARG = eql_v3.float4_ord_ore, RIGHTARG = eql_v3.float4_ord_ore,
+  COMMUTATOR = <@
 );

 CREATE OPERATOR <@ (
   FUNCTION = eql_v3.contained_by,
-  LEFTARG = eql_v3.float4_ord_ore, RIGHTARG = eql_v3.float4_ord_ore
+  LEFTARG = eql_v3.float4_ord_ore, RIGHTARG = eql_v3.float4_ord_ore,
+  COMMUTATOR = @>
 );

Note: This issue exists across all float operator baselines (float4/float8, all variants). Apply the same fix pattern consistently.

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@tests/codegen/reference/float4/float4_ord_ore_operators.sql` around lines 118
- 146, The CREATE OPERATOR statements for @> and <@ operators are missing
COMMUTATOR declarations that define their reciprocal relationships. Add
COMMUTATOR metadata to each operator: for the @> operator with LEFTARG =
eql_v3.float4_ord_ore and RIGHTARG = eql_v3.float4_ord_ore, add COMMUTATOR = <@,
and for the corresponding <@ operator add COMMUTATOR = @>. For the mixed-type
overloads (eql_v3.float4_ord_ore with jsonb), the @> operator with LEFTARG =
eql_v3.float4_ord_ore and RIGHTARG = jsonb should reference the <@ operator with
swapped argument types, and vice versa using the appropriate operator syntax to
specify argument type order. Apply this pattern consistently across all three
pairs of operator definitions in the diff.

🤖 Prompt for all review comments with AI agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Inline comments:
In `@tests/sqlx/src/scalar_domains.rs`:
- Around line 802-815: The F4 struct (and F8 at lines 833-846) derives standard
IEEE PartialEq while implementing Ord using total_cmp, which violates Rust trait
invariants. Remove PartialEq from the derive macro for both F4 and F8 structs,
then implement a custom PartialEq for each that is consistent with their Ord
implementations by using total_cmp for equality comparisons instead of the
default IEEE semantics.

---

Nitpick comments:
In `@tests/codegen/reference/float4/float4_ord_ore_operators.sql`:
- Around line 118-146: The CREATE OPERATOR statements for @> and <@ operators
are missing COMMUTATOR declarations that define their reciprocal relationships.
Add COMMUTATOR metadata to each operator: for the @> operator with LEFTARG =
eql_v3.float4_ord_ore and RIGHTARG = eql_v3.float4_ord_ore, add COMMUTATOR = <@,
and for the corresponding <@ operator add COMMUTATOR = @>. For the mixed-type
overloads (eql_v3.float4_ord_ore with jsonb), the @> operator with LEFTARG =
eql_v3.float4_ord_ore and RIGHTARG = jsonb should reference the <@ operator with
swapped argument types, and vice versa using the appropriate operator syntax to
specify argument type order. Apply this pattern consistently across all three
pairs of operator definitions in the diff.

🪄 Autofix (Beta)

Fix all unresolved CodeRabbit comments on this PR:

• Push a commit to this branch (recommended)
• Create a new PR with the fixes

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Pro

Run ID: 890727e5-db03-4ac5-bd71-2e15200e77c0

📥 Commits

Reviewing files that changed from the base of the PR and between 302a5f1 and c20b8a4.

📒 Files selected for processing (61)

• CHANGELOG.md
• crates/eql-scalars/src/fixture.rs
• crates/eql-scalars/src/kind.rs
• crates/eql-scalars/src/lib.rs
• crates/eql-scalars/src/proptest_invariants.rs
• crates/eql-scalars/src/tests.rs
• crates/eql-tests-macros/src/lib.rs
• crates/eql-types/bindings/v3/Float4.ts
• crates/eql-types/bindings/v3/Float4Eq.ts
• crates/eql-types/bindings/v3/Float4Ord.ts
• crates/eql-types/bindings/v3/Float4OrdOre.ts
• crates/eql-types/bindings/v3/Float8.ts
• crates/eql-types/bindings/v3/Float8Eq.ts
• crates/eql-types/bindings/v3/Float8Ord.ts
• crates/eql-types/bindings/v3/Float8OrdOre.ts
• crates/eql-types/schema/v3/float4.json
• crates/eql-types/schema/v3/float4_eq.json
• crates/eql-types/schema/v3/float4_ord.json
• crates/eql-types/schema/v3/float4_ord_ore.json
• crates/eql-types/schema/v3/float8.json
• crates/eql-types/schema/v3/float8_eq.json
• crates/eql-types/schema/v3/float8_ord.json
• crates/eql-types/schema/v3/float8_ord_ore.json
• crates/eql-types/src/v3/float4.rs
• crates/eql-types/src/v3/float8.rs
• crates/eql-types/src/v3/mod.rs
• docs/plans/2026-06-19-float-encrypted-domain-plan.md
• docs/upgrading/v2.4.md
• tests/codegen/reference/float4/float4_eq_functions.sql
• tests/codegen/reference/float4/float4_eq_operators.sql
• tests/codegen/reference/float4/float4_functions.sql
• tests/codegen/reference/float4/float4_operators.sql
• tests/codegen/reference/float4/float4_ord_aggregates.sql
• tests/codegen/reference/float4/float4_ord_functions.sql
• tests/codegen/reference/float4/float4_ord_operators.sql
• tests/codegen/reference/float4/float4_ord_ore_aggregates.sql
• tests/codegen/reference/float4/float4_ord_ore_functions.sql
• tests/codegen/reference/float4/float4_ord_ore_operators.sql
• tests/codegen/reference/float4/float4_types.sql
• tests/codegen/reference/float8/float8_eq_functions.sql
• tests/codegen/reference/float8/float8_eq_operators.sql
• tests/codegen/reference/float8/float8_functions.sql
• tests/codegen/reference/float8/float8_operators.sql
• tests/codegen/reference/float8/float8_ord_aggregates.sql
• tests/codegen/reference/float8/float8_ord_functions.sql
• tests/codegen/reference/float8/float8_ord_operators.sql
• tests/codegen/reference/float8/float8_ord_ore_aggregates.sql
• tests/codegen/reference/float8/float8_ord_ore_functions.sql
• tests/codegen/reference/float8/float8_ord_ore_operators.sql
• tests/codegen/reference/float8/float8_types.sql
• tests/sqlx/snapshots/matrix_tests_storage_only.txt
• tests/sqlx/src/fixtures/eql_plaintext.rs
• tests/sqlx/src/fixtures/scalar_fixture.rs
• tests/sqlx/src/lib.rs
• tests/sqlx/src/scalar_domains.rs
• tests/sqlx/src/scalar_types.rs
• tests/sqlx/tests/encrypted_domain.rs
• tests/sqlx/tests/encrypted_domain/float_special.rs
• tests/sqlx/tests/encrypted_domain/property/e2e_oracle.rs
• tests/sqlx/tests/encrypted_domain/property/fixture_oracle.rs
• tests/sqlx/tests/encrypted_domain/signed.rs