Skip to content

chore: pin core#93, and prove ! A, B end to end - #67

Merged
Kinflou merged 1 commit into
mainfrom
chore/bump-core-multi-throws
Oct 17, 2026
Merged

Kinflou merged 1 commit into
mainfrom
chore/bump-core-multi-throws

Conversation

@Kinflou

@Kinflou Kinflou commented Oct 17, 2026

Copy link
Copy Markdown
Member

Step 4 of the core#93 cascade. All four pins move together — cli has no
[patch] unifying them, so a partial bump gives it two comline-core copies
and two incompatible FrozenUnit types. comline-codegen is pinned at the rev
both generators pin (f919acf), not at generation's branch tip, which is the
mistake a previous run of this cascade made.

The end-to-end test needed real work

It is the only test anywhere that drives the whole chain: the real binary
parses a schema, generates a crate, and that crate is compiled against
comline-runtime and round-tripped. So it is where the error-handling work
actually gets proven, and the rev bump broke it exactly as expected — the
hand-written provider returned Result<_, ChatSendError> where the trait now
wants Fault<ChatSendError>.

The fixture schema now declares two errors and ! Rejected, TooLong. This is
the first time multi-! is parsed by the actual binary
rather than by a
hand-built IR fixture — core#93's grammar change had until now only been
exercised by core's own tests.

The driver covers:

  • a declared error (Fault::App) decoded back into its own variant
  • the second error of the two, decoded into its own variant and not into
    whichever happens to share an ordinal
  • Display rendered from the schema's message, plain ("rejected: empty")
    and interpolated ("too long: 8 max")
  • an undeclared failure as Fault::Internal, surfacing as
    CallError::Runtime(RuntimeError::Provider) — the escape hatch that did not
    exist before
  • that the connection survives a fault: it is a reply, not a teardown

Verification

92 tests pass, 0 failed. fmt clean.

Note the checkout has an unrelated uncommitted edit to src/commands/new.rs
that is not mine; it is deliberately not part of this commit.

Ordering

playground is the last step and follows this one.

Step 4 of the core#93 cascade — all four pins move together, since the
CLI has no `[patch]` to unify them and a split graph gives two
incompatible `FrozenUnit` types. `comline-codegen` is pinned at the rev
*both generators* pin, not at generation's branch tip.

The end-to-end test needed real work, not just a rev bump. It builds a
generated crate against the runtime and round-trips it, so it is the
only place the whole chain is exercised: the real binary parses the
schema, generates the crate, and the crate talks to the runtime.

The fixture schema now declares two errors and `! Rejected, TooLong`,
which is the first time multi-`!` is parsed by the actual binary rather
than by a hand-built IR fixture. The driver covers:

- a declared error (`Fault::App`), decoded back into its own variant
- the *second* error of the two, decoded into its own variant and not
  into whichever shares an ordinal
- `Display` from the schema's `message` template, both plain
  (`"rejected: empty"`) and interpolated (`"too long: 8 max"`)
- an undeclared failure as `Fault::Internal`, arriving as
  `CallError::Runtime(RuntimeError::Provider)`
- that the connection survives a fault — it is a reply, not a teardown
@Kinflou
Kinflou merged commit b9478af into main Oct 17, 2026
2 checks passed
@Kinflou
Kinflou deleted the chore/bump-core-multi-throws branch November 8, 2026 13:00
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant