Prelude surface: classes, instances, and the value functions

[biuld]

Part of #76 (Phase 3). The highest-leverage library slice: 163 passing files are blocked on Prelude alone, more than every other missing module combined.

stdlib/lib/Prelude.purs is currently six declarations — pure, bind, discard, map, apply, runEffect over foreign import data Effect. The corpus expects the standard classes and instances.

Classes and instances

Show, Eq, Ord, Semigroup, Monoid, Functor, Apply, Applicative, Foldable, plus Boolean combinators, folds, and the String/Char/Number conversions the corpus uses.

Prerequisites this exposes

Each of these is Phase 2 work, and writing the library is the forcing function that finds the rest:

• Instances must resolve — Resolve instance declarations and instance signatures #90. Every class above needs instances, and instances are not in the resolver yet.
• Records and rows on the type spine — class methods take and return records; Show needs Record.toList-shaped display helpers that use open rows.
• do notation with multiple statements and let — already lowered, but the library will stress it.
• Guards — Ord/Eq instance bodies are guard-shaped (Lower guarded equations, guarded case alternatives, and multi-scrutinee case #88).
• Operator sections — (<>), flip, and friends are written as sections (Lower operator, constructor-operator and type-operator aliases #86).

Constraint

Every addition compiles and executes under Wasmtime, not only type-checks. A Prelude that type-checks but cannot run does not recover a single passing file.

Acceptance

Prelude carries the classes and instances above, executes under Wasmtime, and the passing files blocked on Prelude fall toward zero.

[biuld]

First slice landed in #128: Data.Function and the $/# application operators.

What landed

• Data.Function: apply, applyFlipped, const, flip, on, plus infixr 0 apply as $ and infixl 1 applyFlipped as #.
• Prelude re-exports const, flip, $, # (the official list) with an explicit export list.
• Two prerequisite fixes: a re-exported operator alias no longer requires its target's name (the purs rule), and CC's partial-application adapter now converts the parameters it leaves to the caller to the declaration's shapes.

identity is deliberately not defined here: official re-exports it from Control.Category, so it belongs with the class hierarchy rather than as a private Data.Function.identity.

Measured

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

Board	Before	After
passing resolution	209/413	225/413
missing library module	62	61
P3 resolve	138	123
P10	147	155
P5 typecheck	35	40
L1 / M2 / M3 / M4 / M5 / L6	904/908, 72/72, 34/48, 34/50, 51/92, 0/413	unchanged

Board field

The Corpus cases field is now 80: the measured number of passing files whose first blocker is a Prelude class or value name (the convention counts first blocking stage, so this is the latent recovery the issue's surface owns, not a change in the first-blocker total). The largest groups are <> 15, Monad 9, identity/show/Eq 7 each, negate 5, Functor 5, otherwise/&&/<$>/add 3 each, and singles for Semiring, Show, when, append, mod, between, and >>>.

Remaining

The class hierarchy and its instances (Semigroup/<> needs a string- and array-concatenation primitive), Show/logShow (needs number formatting and Data.Show), Eq/Ord and the arithmetic classes whose surface operators are currently Int-monomorphic intrinsics, and identity/Control.Category.

[biuld]

Landed the first slice: Data.Semigroup and <> (PR #130).

• stdlib/lib/Data/Semigroup.purs declares class Semigroup a where append :: a -> a -> a, infixr 5 append as <>, and the String, Unit, and Array a instances; Prelude re-exports class Semigroup, append, <>.
• The compiler gains the arrayAppend primitive (HIR Intrinsic::ArrayAppend → Core ExprKind::ArrayAppend → CC AssignmentKind::ArrayAppend). Because the MIR ArrayNewDefault verifier enforces array-conversion invariants, a general ArrayNewSized allocator plus two element-copy loops lower it.
• Tests: PSRS_REQUIRE_WASMTIME=1 cargo test -p psrs-driver --lib tests::semigroup (4 pass; Wasmtime execution for string/array append including a non-ASCII string, and right-associativity).

Measured with PSRS_ORACLE=annotations cargo test -p psrs-driver --test suite -- --ignored --nocapture (2026-10-04):

Board	Before	After
L2 passing resolution	225/413	235/413
L2 first-stage blockers	188	178
L2 P3 / P10	123 / 155	113 / 160
L6/M7 P5 typecheck / P8 / P6	40 / 11 / 3	43 / 12 / 4
L4	34/50	35/50
L5 NoInstanceFound	38/53	40/53 (aggregate 51/92)

<> was the largest single P3 name (15 cases) and append blocked 2 more; all 17 now stop on a later stage, and 10 cases resolve past L2 (5 P10, 3 P5 typecheck, 1 P8, 1 P6). The board Corpus cases for this issue drops 80 → 70.

Still open here: show/Show, Monad, Eq, identity, Functor/<$>, negate, and the rest of the class/value surface; assertEqual and logShow (shared with #95) still need Eq/Show.

[biuld]

Landed Data.Functor in #142 (merge commit 77c8745).

class Functor, map, and <$> live in Data.Functor, with Array, Maybe, and Either instances. Prelude re-exports them and adds the Effect instance. That instance is the old effect-only map: creating the action does not run it.

One PSRS_REQUIRE_WASMTIME=1 PSRS_ORACLE=annotations scoreboard on 2026-10-04, Wasmtime 49.0.2, purs 0.15.16:

Board	Before	After
passing resolution	270/413	276/413
L3	35/48	36/48
L4	35/50	35/50
L5	53/80	53/79
L6/M7	124/413	125/413

All 125 runtime files exit 0. The added file is passing/3549.purs. failing/3549.purs now matches KindsDoNotUnify. failing/NewtypeInstance5.purs leaves the L5 board and fails in the newtype checker. L1 stays 904/908 and L2 failing stays 72/72.

Functor and <$> are gone from the first-blocker names. Corpus cases is now 67: Monad 14, identity 9, _ 8, negate 6, Eq1 4, <<< 4, otherwise 3, && 3, when 2, mod 2, compare 2, and one each of between, P.identity, >>>, <*>, not, Apply, ||, Applicative, -, and Foo.Bar.

#94 stays open.