Effect, Effect.Console and Test.Assert surfaces over the WASI library

[biuld]

Part of #76 (Phase 3). Remeasured 2026-10-03 on master at fe40d17: 258 passing files stop at Effect.Console, 43 at Effect, and 4 at Test.Assert — 305 files, by first blocking stage. No passing file blocks on Prelude, because stdlib/lib/Prelude.purs exists and resolves; its missing surface is latent rather than a first blocker, which is why #94 measures 0 by this convention.

We already have the runtime: WASI.Console wraps wasi:cli/stdout/stderr, and WASI-02/03 are Verified with execution tests. What is missing is the module surface the corpus expects.

Do

• Effect and Effect.Console as a source-level surface over the existing WASI console, exporting log, logShow, warn, error and the Effect type the corpus imports.
• Keep WASI.Console as the platform layer and have Effect.Console be a thin binding over it, so the WASI capability gates stay in one place.
• Test.Assert over the same execution model: assert, assertEqual, assertThrows, assertFailure. A failing assertion must be observable as a non-zero exit, so passing assertion files score cleanly.

Note on DEC-11

docs/decision/DEC-11-primitive-ffi-stdlib-wrappers.md fixes the wrapper-versus-raw-import split. Effect.Console is a wrapper; the raw import stays inside WASI.Console. Do not collapse them.

Acceptance

Effect, Effect.Console, and Test.Assert exist, execute under Wasmtime, and the 305 blocked files drop toward zero.

[biuld]

Second slice landed in #127: the unit value and Test.Assert.

What landed

• unit as a compiler primitive (Intrinsic::Unit). It lowers to Core ExprKind::Unit and to the canonical integer 0 only in CC, so no earlier stage reads it as an Int. unit and () reach the same value.
• Test.Assert with assert, assert', assertTrue, assertFalse.
• Prelude.trap, a psrs:effect external whose body is an unreachable path. A failed assertion writes its message to stderr and then traps; a guest trap, not a non-zero exit code, is the only failure signal a compilable corpus case can carry.

PSRS_REQUIRE_WASMTIME=1 cargo test -p psrs-driver --lib tests::assertions executes both paths (held assertion exits 0; failed one writes first, then traps, and nothing after the trap runs).

Measured (PSRS_ORACLE=annotations cargo test -p psrs-driver --test suite -- --ignored --nocapture)

Board	Before this slice	After
passing resolution	199/413	209/413
missing library module	83	62
P3 resolve	127	138
L1 / L3 / L4 / L5 / L6	904/908, 34/48, 34/50, 51/92, 0/413	unchanged

The 21 Test.Assert cases split into 10 that now resolve and 11 that stop on the next library gap; the unit half clears 11 unknown name unit blockers. The board's Corpus cases for this issue is now 18.

What remains on this issue

logShow (16 cases) and assertEqual (2 cases), both waiting on the Show/Eq class surface that #94 owns. assertThrows needs to observe a trap, which the target profile does not provide, so it is recorded as an obligation rather than approximated (D-04 M7, EF-12). The issue stays open until #94 lands that surface.

[biuld]

Effect.Console.logShow landed — #138

logShow :: forall a. Show a => a -> Effect Unit is now defined as
log (show value) over the Show class from Data.Show. It adds no I/O and no
second stringifier: the rendering is the library's Show and the destination is
still the one WASI.Console.log decides. Verified under Wasmtime by
tests::assertions::log_show_writes_the_library_rendering, which asserts the
renderings for Int, a negative Int, String, Char, Boolean, and
Array Int.

Evidence

PSRS_REQUIRE_WASMTIME=1 PSRS_ORACLE=annotations cargo test -p psrs-driver --test suite -- --ignored --nocapture

at 071fb11 plus this slice, one run of all five boards:

Board	Before	After
L1 parse	904/908	904/908
L2 failing	72/72	72/72
L3	35/48	35/48
L4	35/50	35/50
L5	53/80	53/80
passing resolved	253/413	270/413
M7 first blockers	missing 53, P10 177, P3 103, P5t 42, P5k 17, P8 13, P6 4, P0 4	missing 53, P10 187, P3 86, P5t 46, P5k 17, P8 14, P6 6, P0 4
L6/M7 runtime	0/413	0/413

logShow was the largest single P3 name blocker at 24 cases, 19 of them reported
as "logShow is not exported by the imported module". 17 resolve past L2; the
other 7 stop at the next P3 name, which nothing shadowed before — Monad and
compare 2 each, and negate, not, and _ 1 each. The 17 land at later stages
as 10 P10, 4 P5 typecheck, 1 P8, and 2 P6.

Board Corpus cases for this issue is now 2.

assertEqual/assertEqual' are not landed, and the reason changed

Both classes are declared, so this is no longer a library-surface gap — it is a
compiler defect, filed as #137.

The corpus calls the record form (passing/DerivingFoldable.purs and one other
case):

assertEqual' :: forall a. Eq a => Show a => String -> { actual :: a, expected :: a } -> Effect Unit

A constraint whose quantified variable appears inside a record type is
elaborated with the record as the constraint's argument, so the wanted
constraint comes out as Eq { actual :: a, expected :: a } and the declaration is
rejected with NoInstanceFound. Probes that separate this from record syntax:

Signature	Result
forall a. Eq a => a -> Int	accepted
forall a. String -> { actual :: a, expected :: a } -> Int (no constraint)	accepted
forall a. { actual :: a, expected :: a } -> a — field access	accepted
forall a. { actual :: a, expected :: a } -> a — record pattern	accepted
forall a. Eq a => { actual :: a, expected :: a } -> Int	rejected
the same behind a type synonym for the record	rejected

A record whose fields are quantified variables is fine on its own, and both
destructuring forms work on it; only the constraint combined with the record
fails, and a type synonym does not change it, so this is constraint elaboration
rather than record syntax or the parser. Changing the argument shape would change
the API the corpus calls, so stdlib/lib/Test/Assert.purs records the blocker
instead of approximating it.

Still open on this issue

assertEqual and assertEqual' (2 passing cases), blocked on #137.
assertThrows/assertThrows' remain a separate gap: they need to observe that an
argument trapped, which needs the Wasm exceptions proposal, outside DEC-05's
profile.

Status stays In Progress.