jolt/docs/spec/09-core-library.md
Yogthos f80f9aab4b One IRef seam: watches/validators/meta over atom, var, and agent
add-watch/remove-watch/set-validator!/get-validator were atom-only; the
atom ctor ignored :meta and :validator; watching a var crashed. Now the
ARef contract is one seam: atoms keep their record slots (hot path
unchanged), every other reference type registers a predicate and stores
watches/validators in identity-keyed side tables, and notifies at its
mutation points. Vars notify on root changes (def on a watched var,
var-set outside a thread binding, alter-var-root — thread-binding sets
don't notify, like the JVM); agents notify per action. The def-var! wrap
costs two weak-table probes per def and does IRef work only on a watched
var.

Ctor options follow ARef: the validator gates the initial value
(IllegalStateException 'Invalid reference state' — also the class for
rejected swap!/reset!), :meta must be a map (else ClassCastException),
nil allowed. meta reads any reference through the identity side-table
(the type-gated fall-through is gone); alter-meta!/reset-meta! work on
non-var references.

Runtime-only (no re-mint). 9 JVM-certified corpus rows; spec entry; cts
baseline 5781 -> 5805 pass, 73 baselined namespaces (the residual error
in the watch namespaces is their STM ref section — refs stay out of
scope).
2026-07-02 09:07:00 -04:00

333 lines
13 KiB
Markdown
Raw Permalink Blame History

This file contains ambiguous Unicode characters

This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.

# §9 The Core Library
**Status**: entry format fixed; exemplars for `first`, `reduce`, `parse-uuid`.
The full portable surface (≈500 vars after classification, dashboard in
`coverage.md`) is filled in chapter-by-chapter using this format.
Entries specify *behavioral contracts*, not implementations. Performance
characteristics are specified only where the language community relies on
them (e.g. vector `nth` is "effectively constant time" — SHOULD-level).
---
## Collection return types & laziness (cross-cutting)
Two contracts hold across the sequence library and are not restated per entry.
**Return-type fidelity.** A function returns the same *kind* of collection the
reference does — value equality is not enough, since `(= [0 1] '(0 1))`.
- Sequence transformations return **seqs** (lazy unless noted): `map`, `filter`,
`remove`, `keep`, `mapcat`, `take`/`drop` and their `-while` forms, `partition`,
`partition-all`, `partition-by`, `interpose`, `dedupe`, `distinct`, `concat`,
`reductions`, `cons`, `rest`, `sequence`. The *elements* of `partition` /
`partition-all` / `partition-by` are themselves seqs, not vectors.
- The vector variants return **vectors**: `mapv`, `filterv`, `vec`, `subvec`,
`partitionv`, `partitionv-all`, `splitv-at`. `split-at` / `split-with` return a
2-vector `[take drop]`. A transducer applied eagerly (`into []`, the
`partition-all` transducer's chunks) yields vectors.
- Type-preserving functions return the input's type: `replace` over a vector is a
vector, over any other seqable a (lazy) seq; `empty`/`into (empty coll)` keep the
collection kind; `set`/`into #{}` return sets; `into {}`/`select-keys`/`zipmap`/
`frequencies`/`group-by`/`merge` return maps (`group-by` values are vectors).
**Laziness.** The lazy sequence functions — including `sequence`, `eduction`, and
`mapcat` — MUST consume their source incrementally and so terminate on an infinite
or unbounded source when only a prefix is demanded: `(first (sequence (map inc)
(range)))` and `(take n (mapcat f (range)))` return without realizing the whole
source. `(apply concat coll-of-colls)` is likewise lazy in its argument seq. The
eager consumers (`reduce`, `into`, `count`, `vec`, `doall`) realize the demanded
portion fully.
These are exercised by the `seq / lazy over infinite` and the per-fn type-predicate
rows in the conformance corpus.
---
### first — since 1.0
```
(first coll)
```
**Semantics**
- S1. MUST return the first element of `(seq coll)`.
- S2. If `(seq coll)` is `nil` (i.e. `coll` is empty or `nil`), MUST return
`nil`.
- S3. MUST accept anything *seqable* (§5): seqs, lists, vectors, maps
(yielding map entries), sets, strings (yielding characters), `nil`.
- S4. On a lazy sequence, MUST realize at most the first element (§5
laziness contract).
**Edge cases**
- E1. `(first nil)``nil`; `(first [])``nil`; `(first "")``nil`.
- E2. A `nil` or `false` first *element* is returned as-is — callers cannot
distinguish "empty" from "first element is nil" via `first` alone (that is
what `seq` is for).
- E3. On a map, the element is a map entry; on an unordered collection (map,
set) *which* element is first is implementation-defined but MUST be
consistent with that collection's seq order for the same collection value.
**Errors**
- X1. A non-seqable argument (e.g. a number) MUST throw.
**Examples**
```clojure
(first [1 2 3]) ;=> 1
(first '()) ;=> nil
(first "ab") ;=> \a
(first {:a 1}) ;=> [:a 1]
(first [nil 2]) ;=> nil
```
**Conformance**
S1S3, E1E2 → jolt `sequences-spec` "seq / access"; clojure-test-suite
`core_test/first.cljc`. S4 → jolt `lazy-seqs-spec` counter cases. X1 →
clojure-test-suite `core_test/first.cljc` (throwing cases).
---
### reduce — since 1.0
```
(reduce f coll)
(reduce f init coll)
```
**Semantics**
- S1. With `init`: MUST return `init` if `(seq coll)` is nil; otherwise MUST
return `(f … (f (f init e₁) e₂) … eₙ)`, applying `f` left-to-right over the
elements, exactly once each.
- S2. Without `init`: if `coll` is empty, MUST return `(f)` (f called with
no arguments); if `coll` has one element, MUST return that element
*without calling `f`*; otherwise as S1 with `init = e₁` over `e₂…eₙ`.
- S3. **Reduced short-circuit**: if any intermediate result is a `reduced`
value, iteration MUST stop and the dereferenced value MUST be returned
immediately; `f` MUST NOT be called again.
- S4. `reduce` is eager: it MUST fully realize the consumed portion of a
lazy `coll` (to the end, or to the `reduced` point).
**Edge cases**
- E1. `(reduce f nil)``(f)`; `(reduce f init nil)``init`.
- E2. A `reduced` value as the *initial* `init` is NOT unwrapped before the
first call in the reference — ⚠ under-documented; differential result to
pin down and test before this entry is marked verified.
- E3. Visit order over maps is entry order of the map's seq;
over vectors/lists/seqs it is sequential order (normative).
**Errors**
- X1. Without `init`, on an empty coll, if `f` has no zero-arg arity the
call `(f)` MUST throw (arity error).
**Examples**
```clojure
(reduce + [1 2 3 4]) ;=> 10
(reduce + 10 [1 2 3 4]) ;=> 20
(reduce + []) ;=> 0 ; (+) is 0
(reduce + [5]) ;=> 5 ; f not called
(reduce (fn [a x] (if (> a 2) (reduced a) (+ a x))) 0 [1 2 3 4 5]) ;=> 3
```
**Conformance**
S1S3, E1 → jolt `sequences-spec` "map filter reduce" group +
`transducers-spec` "reduce honors reduced"; clojure-test-suite
`core_test/reduce.cljc`. S2 (single-element, f-not-called) → jolt conformance
"reduce single no init". E2 → UNVERIFIED (differential test to add). S4 →
`lazy-seqs-spec`.
---
### parse-uuid — since 1.11
```
(parse-uuid s)
```
**Semantics**
- S1. If `s` is a string in canonical UUID form — five groups of hex digits
of lengths 8, 4, 4, 4, 12 separated by `-` — MUST return a UUID value `u`
such that `(uuid? u)` is true and `(str u)` is the lowercase form of `s`.
- S2. Parsing MUST be case-insensitive and equality on the results
case-insensitive: `(= (parse-uuid s) (parse-uuid (upper-case s)))` is true.
- S3. If `s` is a string not in canonical form, MUST return `nil`.
⚠ reference-divergence: reference Clojure (java.util.UUID) additionally
accepts non-canonical forms like `"0-0-0-0-0"`; ClojureScript and other
dialects are strict. This spec adopts **strict** (the cross-dialect
behavior); the reference's permissiveness is recorded as host leniency.
- S4. UUID values MUST support value equality, hashing (usable as map keys
and set members), `str` (lowercase canonical form), and print as the
tagged literal `#uuid "…"` such that the printed form reads back equal
(§2 tagged literals).
**Edge cases**
- E1. `""`, over-long, truncated, non-hex characters, and misplaced dashes
`nil`.
**Errors**
- X1. A non-string argument MUST throw.
**Examples**
```clojure
(parse-uuid "b6883c0a-0342-4007-9966-bc2dfa6b109e") ;=> #uuid "b6883c0a-…"
(uuid? *1) ;=> true
(parse-uuid "df0993") ;=> nil
(parse-uuid 1000) ;; throws
```
**Conformance**
S1S4, E1, X1 → jolt `uuid-spec` (30 cases) + 6 three-path conformance
cases; clojure-test-suite `core_test/parse_uuid.cljc`,
`core_test/uuid_qmark.cljc`, `core_test/random_uuid.cljc`.
---
### clojure.template/apply-template, clojure.test/are — since 1.1
```
(apply-template argv expr values)
(are argv expr & args)
```
**Semantics**
- S1. `apply-template` MUST replace every occurrence of each `argv` symbol
in `expr` with its corresponding value by structural walk (postwalk symbol
substitution), not by lexical binding. Occurrences inside `quote` and at
any nesting depth substitute: `(apply-template '[x] '(f 'x) '[if])`
`(f 'if)`.
- S2. `do-template` MUST partition `args` by `(count argv)` and expand to a
`do` of one substituted `expr` per group.
- S3. `clojure.test/are` MUST expand through `do-template` with `expr`
wrapped in `is`. Consequently `(are [x] (special-symbol? 'x) if def)`
asserts `(special-symbol? 'if)` and `(special-symbol? 'def)` — a
let-binding implementation is non-conforming (the quoted symbol would not
substitute).
**Errors**
- X1. `are` MUST throw at macroexpansion when `(count args)` is not a
positive multiple of a non-empty `(count argv)` (empty/empty is allowed).
- X2. `apply-template` MUST throw when `argv` is not a vector of symbols.
**Conformance**
S1S3 → `test/chez/clojure-test.clj` (are with quoted template var);
clojure-test-suite `core_test/special_symbol_qmark.cljc` and every
`are`-based suite namespace.
---
### make-hierarchy, derive, underive, isa?, parents, ancestors, descendants — since 1.0
```
(make-hierarchy)
(derive tag parent) (derive h tag parent)
(underive tag parent) (underive h tag parent)
(isa? child parent) (isa? h child parent)
(parents tag) (ancestors tag) (descendants tag) ; + (f h tag) forms
```
**Semantics**
- S1. A hierarchy is a pure value `{:parents {tag #{...}} :ancestors {...}
:descendants {...}}`; the 3-arity forms are pure, the shorter arities read and
mutate the global hierarchy.
- S2. `isa?` is true when `(= child parent)`, when the host type system says
parent is assignable from child (both classes), when the relationship was
`derive`d — including a relationship derived on one of a class child's
supers — or component-wise for equal-length vectors.
- S3. Class tags answer through the host type hierarchy: `(parents c)` includes
the class's direct supers (`bases` — a concrete class's chain roots at
`java.lang.Object`, an interface's does not); `(ancestors c)` is the
transitive set plus anything `derive`d on the class or its supers. A
deftype/defrecord class's ancestry includes its implemented protocol
interfaces and, for records, the record interfaces
(`clojure.lang.IRecord`/`IPersistentMap`/`Associative`/…; `clojure.lang.IType`
for a bare deftype).
- S4. `derive` returns the updated hierarchy (3-arity) or nil (2-arity);
deriving a relationship that already holds transitively, or one that would
create a cycle, throws.
**Errors**
- X1. `derive` asserts its argument shapes: parent must be a namespaced Named
value; tag must be a class or a Named value (namespaced in the 2-arity
global form); `(derive h tag tag)` fails the `not=` assert. AssertionError.
- X2. `underive`/`derive` with a non-hierarchy `h` throw at the parents
lookup (the map is called as a function, like the reference).
- X3. `(descendants h SomeClass)` throws UnsupportedOperationException
("Can't get descendants of classes") — Java type inheritance is not
enumerable downward.
**Conformance**
S1S4, X1X3 → corpus `hierarchy / *` rows; clojure-test-suite
`core_test/{derive,underive,isa_…,parents,ancestors,descendants}.cljc`
(all fully passing).
---
### atom, add-watch, remove-watch, set-validator!, get-validator — since 1.0
```
(atom x & {:keys [meta validator]})
(add-watch iref key f) (remove-watch iref key)
(set-validator! iref f) (get-validator iref)
```
**Semantics**
- S1. Watches, validators, and reference metadata are one contract (the JVM's
ARef/IRef) shared by atoms, vars, and agents. `add-watch`/`remove-watch`
return the reference; re-adding a key replaces that watch in place.
- S2. A watch is called `(f key ref old new)` after a state change: atom
swap!/reset!/compare-and-set!, var ROOT changes (`def` on a watched var,
`var-set` outside a thread binding, `alter-var-root` — a thread-binding set
does not notify), and each agent action's state change.
- S3. A validator gates every state change and, via the `:validator` ctor
option, the initial value — an invalid initial value never constructs the
reference.
- S4. The `:meta` ctor option attaches reference metadata (`meta` reads it,
`alter-meta!`/`reset-meta!` update it); nil is allowed.
**Errors**
- X1. A rejected value (validator returns logical false or the ctor option
fails on the initial value) throws IllegalStateException "Invalid reference
state".
- X2. A non-map `:meta` ctor option throws ClassCastException.
**Conformance**
S1S4, X1X2 → corpus `iref / *` rows; clojure-test-suite
`core_test/{atom,add-watch,remove-watch}.cljc` (the remaining baselined error
in the watch namespaces is their STM `ref` section — refs are out of scope,
`stm-refs` in `coverage.md`).
---
## Authoring notes
- Source examples from the ClojureDocs export (`clojuredocs-export.edn`,
648 core vars have community examples) — but every example is verified
against the reference before inclusion.
- When writing an entry surfaces a behavior question, settle it by
differential test first; if dialects split, that's a classification
decision (host-dependent / divergence note), not a coin flip.
- An entry is **Verified** when no field carries UNVERIFIED; `coverage.md`
tracks per-var status.