Skip to content

chore(release): v0.4.3 #200

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
avrabe merged 3 commits into from
Apr 23, 2026
+783 −4
Merged

chore(release): v0.4.3 #200

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
c43a59e
feat(variant): rivet variant explain for debugging solve outcomes
avrabe Apr 23, 2026
6cf1c41
test(variant): enrich eu-adas-c example + per-format smoke on realist…
avrabe Apr 23, 2026
06f10b9
chore(release): v0.4.3
avrabe Apr 23, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
90 changes: 90 additions & 0 deletions CHANGELOG.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -5,6 +5,96 @@

## [Unreleased]

## [0.4.3] — 2026-04-23

### `rivet variant` — build-system query surface and solve debugger

Three new subcommands complete the variant-scoped CLI surface
(`REQ-046`). Feature models can now carry typed `attributes:` per
feature, round-tripped through `solve()` and emitted into seven
different build systems — the same one variant YAML can configure
Cargo, CMake, Bazel, a C/C++ header, Make, shell env, or structured
JSON without divergent hand-written shims.

- `rivet variant features --format {json,env,cargo,cmake,cpp-header,bazel,make}`
emits every effective feature plus its `attributes:` entries with long,
namespaced identifiers (`RIVET_FEATURE_*`, `RIVET_ATTR_*`). Every format
is **loud on failure** — a variant that violates a constraint exits
non-zero with the violation list, never a partial emission.
Non-scalar attribute values (lists/maps) only serialise through
`--format json`; build-system formatters return `Error::Schema` rather
than invent a silent flattening convention.

- `rivet variant value FEATURE` — shell-friendly single-feature probe with
exit codes `0` (selected), `1` (unselected), `2` (unknown feature or
variant fails to solve). Designed for `if rivet variant value … ; then …`.

- `rivet variant attr FEATURE KEY` — print one attribute value. Scalars
print bare; list/map values print as JSON so shells can parse
structurally.

- `rivet variant explain [FEATURE]` — dev/debug UX for "why did my
variant pick/skip feature X?". Full audit mode prints every effective
feature with its origin (`selected` / `mandatory` / `implied by <X>` /
`allowed`), plus the unselected set and the full constraint list.
Single-feature focus mode zooms on one feature and lists every
constraint that mentions it.

Feature models gained an `attributes:` key per feature, parsed as
`BTreeMap<String, serde_yaml::Value>`. The shipped
`examples/variant/feature-model.yaml` now carries realistic metadata
(`asil-numeric`, `compliance`, `locale`) so the worked examples in
`docs/getting-started.md` run against the fixture and produce the
documented output.

Test coverage: 11 unit tests in `rivet_core::variant_emit::tests` for
per-format rendering, 15 integration tests in
`rivet-cli/tests/variant_emit.rs` for CLI end-to-end, exit-code
contract, loud-on-failure path, and the realistic-example smoke across
all seven formats.

### S-expression follow-ups

- `(> (count <scope>) N)` now lowers to a new `CountCompare` expr
variant that evaluates the count against the store once and compares
to an integer threshold. Previously the audit documented `(count …)`
as "meant for numeric comparisons" but no lowering existed — you
could only use it as a standalone predicate. Every comparison operator
(`>`, `<`, `>=`, `<=`, `=`, `!=`) now accepts a `(count …)` LHS with
an integer RHS.

- `(matches <field> "<regex>")` validates the regex at lower time
instead of silently returning `false` at runtime on malformed
patterns. Closes the "mysterious empty result" footgun — typing
`(matches id "[")` used to match nothing and cost debug time; now it
produces a parse error with the compiler's message. Non-literal
patterns (rare; from field interpolation) still use the runtime-lenient
path.

- `docs/getting-started.md` gains dedicated sections for count
comparisons and regex validation, plus a note that dotted accessors
like `links.satisfies.target` are not supported — use the purpose-built
`linked-by` / `linked-from` / `linked-to` / `links-count` predicates.

### Rivet Delta CI action — SVG render for email/mobile

`rivet-delta.yml` workflow now pre-renders the summary Mermaid diagram
to SVG and pushes it to an orphan `rivet-delta-renders` branch, so email
notifications and the GitHub mobile app show the diagram inline instead
of a `<mermaid>` text block that nothing except the web UI can render.
Classification-priority ordering in `scripts/diff-to-markdown.mjs` is
also fixed so multi-label changes (`breaking` + `additive` → `breaking`)
pick the most severe.

### Stamp command

- `rivet stamp all --missing-provenance` filter now correctly checks the
first-class `provenance:` struct field (previously it looked for a
`provenance` entry in generic `fields:` and was therefore a no-op).
- `set_provenance` no longer aborts the whole batch on a single
CST-invisible artifact; it warns and skips that one artifact and
continues.

### Safety-Critical Rust Consortium (SCRC) clippy escalation — Phase 1

Follow-up to the v0.4.2 commitment recorded in `DD-058`. The full
Expand Down
6 changes: 3 additions & 3 deletions Cargo.lock
View file
Open in desktop

Some generated files are not rendered by default. Learn more about how customized files appear on GitHub.

2 changes: 1 addition & 1 deletion Cargo.toml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -7,7 +7,7 @@ members = [
]

[workspace.package]
version = "0.4.2"
version = "0.4.3"
authors = ["PulseEngine <https://github.com/pulseengine>"]
edition = "2024"
license = "Apache-2.0"
Expand Down
262 changes: 262 additions & 0 deletions artifacts/v043-artifacts.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -198,3 +198,265 @@ artifacts:
created-by: ai-assisted
model: claude-opus-4-7
timestamp: 2026-04-22T21:30:00Z

# ── Variant query surface (v0.4.3 headline feature) ─────────────────

- id: DD-061
type: design-decision
title: Build-system emitter formats are namespaced and loud-on-failure
status: approved
description: >
`rivet variant features` emits long, namespaced identifiers
(`RIVET_FEATURE_<NAME>`, `RIVET_ATTR_<FEATURE>_<KEY>`) in every
format so a downstream project can embed several rivet models
without collision. Every formatter is loud on two conditions:
(a) a variant that violates a constraint exits non-zero with the
violation list, never a partial emission; (b) a non-scalar
attribute value (list or map) is only preserved by `--format json`
— the build-system formatters return `Error::Schema` rather than
invent a silent flattening convention. Both are explicit
rejections of the "silent accept" antipattern this project
tracks in `REQ-004`.
tags: [variant, cli, loud, design-decision]
links:
- type: satisfies
target: REQ-046
- type: depends-on
target: DD-050
fields:
rationale: >
The user's v0.4.3 direction was explicit: "1 Rust Cargo Bazel
cmake c cpp and generics, 2 both [boolean+string attributes],
3 loud, 4 long [namespace]". Namespace-prefixed names win on
coexistence; loud-on-failure wins on debuggability; loud-on-
non-scalar wins on user control (they decide the flattening,
the tool does not guess). The alternative — silent flattening
by comma-join or last-write-wins — was rejected because the
attribute YAML is the user's source of truth, and making the
tool pick a convention invisibly would have been the same
footgun as the parse-time validations we added in #196.
alternatives-considered: >
(1) Short identifiers (FEATURE_ASIL_C) — rejected because it
conflicts with hand-rolled feature flags many projects already
have in their build system. (2) Silent flattening of
non-scalars — rejected, see rationale. (3) One format per
subcommand (`rivet variant cargo`) — rejected, single flag
keeps the CLI surface narrow and predictable.
provenance:
created-by: ai-assisted
model: claude-opus-4-7
timestamp: 2026-04-23T05:25:00Z

- id: FEAT-130
type: feature
title: rivet variant features/value/attr — build-system query surface
status: approved
description: >
Three new subcommands on the variant-scoped CLI surface. `rivet
variant features --format {json,env,cargo,cmake,cpp-header,
bazel,make}` emits the resolved feature set plus per-feature
`attributes:` entries in seven build-system-specific formats.
`rivet variant value FEATURE` is a shell-friendly probe with
exit codes 0/1/2 (on/off/unknown). `rivet variant attr FEATURE
KEY` prints a single attribute. Feature models gained an
`attributes:` map per feature, parsed as `BTreeMap<String,
serde_yaml::Value>` and round-tripped through `solve()`.
tags: [variant, cli, build-system, v043]
links:
- type: implements
target: REQ-046
- type: satisfies
target: DD-061
fields:
source-ref: >
rivet-core/src/variant_emit.rs (formatters + unit tests)
rivet-core/src/feature_model.rs (attributes field)
rivet-cli/src/main.rs (cmd_variant_features/value/attr)
rivet-cli/tests/variant_emit.rs (CLI integration tests)
examples/variant/feature-model.yaml (enriched with realistic
asil-numeric, compliance, locale attributes)
verification: >
11 unit tests in rivet_core::variant_emit::tests for per-format
rendering (slug rules, sh-quoting, loud-error path, JSON
structure preservation).
15 integration tests in rivet-cli/tests/variant_emit.rs
covering CLI end-to-end, exit-code contract, loud-on-failure
path, and a smoke run against the shipped eu-adas-c example.
Full cargo test --workspace green.
provenance:
created-by: ai-assisted
model: claude-opus-4-7
timestamp: 2026-04-23T05:25:00Z

- id: FEAT-131
type: feature
title: rivet variant explain — debug why the solver picked what it picked
status: approved
description: >
`rivet variant explain [FEATURE]` answers the "why did my
variant pick/skip feature X?" question. Full audit mode (no
FEATURE) prints every effective feature with its origin
(`selected` / `mandatory` / `implied by <X>` / `allowed`),
plus the unselected set and the constraint list. Single-feature
focus mode prints one feature's state, origin, attribute values,
and every constraint that mentions it. `--format json` emits
the same structured audit for scripts.
tags: [variant, cli, dev-ux, debugging, v043]
links:
- type: implements
target: REQ-046
- type: satisfies
target: DD-050
fields:
rationale: >
The user called this out explicitly: "we should do in parallel
that we are really good on developing and debugging this
stuff". The solver already tracks FeatureOrigin per effective
feature; the missing piece was surfacing it on the CLI. This
lands that surface.
source-ref: >
rivet-cli/src/main.rs (cmd_variant_explain, VariantAction::Explain).
3 integration tests covering text+JSON modes and full-variant
audit.
provenance:
created-by: ai-assisted
model: claude-opus-4-7
timestamp: 2026-04-23T05:25:00Z

# ── S-expression hardening ──────────────────────────────────────────

- id: DD-062
type: design-decision
title: matches regex and count-compare validated at lower time, not runtime
status: approved
description: >
Regex patterns in `(matches <field> "<regex>")` and the integer
RHS of a `(count …)` comparison are validated when the filter
expression is lowered from s-expression to `Expr`, not lazily at
evaluation time. A malformed regex or a non-integer count
threshold produces a parse error with the compiler's message,
not a silent empty result set.
tags: [sexpr, validation, loud, design-decision]
links:
- type: satisfies
target: REQ-004
- type: depends-on
target: DD-058
fields:
rationale: >
The sexpr audit (PR #194) documented `(count …)` as "meant
for numeric comparisons" but no lowering existed. Users
typing `(> (count …) 10)` saw silent no-match. Separately,
`(matches id "[")` compiled but never matched — the regex
crate failed at runtime and the evaluator swallowed the
error. Both are classic silent-accept footguns. Validating
at lower time converts the footgun into a build-time error
and matches the broader "loud" direction the project has
adopted across v0.4.2 and v0.4.3.
provenance:
created-by: ai-assisted
model: claude-opus-4-7
timestamp: 2026-04-23T05:25:00Z

- id: FEAT-132
type: feature
title: count-compare lowering + matches parse-time regex validation
status: approved
description: >
New `Expr::CountCompare(Box<Expr>, CompOp, i64)` variant. The
lowering pass detects a `(count <scope>)` on the LHS of any of
the six comparison operators (>, <, >=, <=, =, !=) with an
integer literal on the RHS and produces CountCompare, which the
evaluator realises by counting matching artifacts once and
comparing. Separately, `(matches <field> "<literal-regex>")`
now calls `regex::Regex::new` at lower time; a parse error
surfaces with the compiler's message. Non-literal patterns (rare;
field interpolation) still use the runtime-lenient path.
tags: [sexpr, lowering, validation, v043]
links:
- type: implements
target: REQ-004
- type: satisfies
target: DD-062
fields:
source-ref: >
rivet-core/src/sexpr_eval.rs — CountCompare variant,
count-extract helpers, regex parse-time validation.
rivet-core/tests/sexpr_predicate_matrix.rs — renamed test
pins new strict behavior.
rivet-core/tests/sexpr_fuzz.rs — expr_to_sexpr handles the
new variant so fuzz roundtrip stays equivalent.
verification: >
5 new regression tests: count_compare_gt_threshold,
count_compare_requires_integer_rhs,
count_compare_all_six_operators_lower,
matches_rejects_invalid_regex_at_lower_time,
matches_accepts_valid_regex.
provenance:
created-by: ai-assisted
model: claude-opus-4-7
timestamp: 2026-04-23T05:25:00Z

# ── CI / release pipeline ───────────────────────────────────────────

- id: FEAT-133
type: feature
title: Rivet Delta — pre-rendered Mermaid SVG for email/mobile viewers
status: approved
description: >
The rivet-delta GitHub Action now pre-renders the summary
Mermaid diagram to SVG via `@mermaid-js/mermaid-cli` and pushes
the rendered asset to an orphan `rivet-delta-renders` branch at
`pr-<N>/run-<N>/diagram.svg`. The PR comment embeds both the
Mermaid source (web UI) and an `<img src=...>` pointing at the
SVG so email notifications and the GitHub mobile app show the
diagram inline. Classification-priority ordering in
`scripts/diff-to-markdown.mjs` is also fixed so multi-label
changes (breaking+additive → breaking) pick the most severe.
tags: [ci, rivet-delta, ux, v043]
links:
- type: traces-to
target: FEAT-010
fields:
source-ref: >
.github/workflows/rivet-delta.yml — SVG render step + orphan
branch push.
scripts/diff-to-markdown.mjs — --mmd-out, --svg-url flags,
classification priority.
verification: >
Playwright e2e test renders a fixture delta and asserts both
the Mermaid block and the SVG anchor appear in the comment
body. ts-transpile smoke test runs against the spec file.
provenance:
created-by: ai-assisted
model: claude-opus-4-7
timestamp: 2026-04-23T05:25:00Z

- id: FEAT-134
type: feature
title: rivet stamp — correct --missing-provenance filter + warn-skip
status: approved
description: >
`rivet stamp all --missing-provenance` previously checked for a
`provenance` entry in generic `fields:`, which was always
absent because provenance is a first-class struct field on
Artifact. The filter now correctly checks `a.provenance.is_none()`.
Separately, `set_provenance` no longer aborts the whole batch on
a single CST-invisible artifact; it warns and skips that item
and continues stamping the rest.
tags: [stamp, cli, fix, v043]
links:
- type: verifies
target: REQ-004
fields:
source-ref: >
rivet-cli/src/main.rs — cmd_stamp_all filter predicate.
rivet-core/src/store.rs — set_provenance warn-skip path.
verification: >
stamp integration tests cover both the filter (an artifact
with no provenance is stamped) and the warn-skip path
(mixing CST-visible + CST-invisible items completes).
provenance:
created-by: ai-assisted
model: claude-opus-4-7
timestamp: 2026-04-23T05:25:00Z
Loading
Loading