solve-engine 2.15.0 → 2.17.0

package/dist/{TokenNormalizer-BfTHg1qt.d.cts → TokenNormalizer-MXaKLJ_m.d.cts}

@@ -69,6 +69,43 @@ interface NormalizerMatch {
      */
     ruleName?: string;
 }
+/**
+ * One position of a rule's leading shape, as a declarative constraint.
+ *
+ * A slot says what the token at that offset from the match position may be.
+ * Both fields are optional and an omitted one constrains nothing, so `{}` is a
+ * wildcard slot, useful for reaching past a position a rule does not care about
+ * to one it does.
+ *
+ * The exactness contract runs in one direction only, and it is the whole reason
+ * this is safe to adopt gradually. Declaring MORE than the rule can match costs
+ * a `match()` call that returns null, which is what happens today anyway.
+ * Declaring LESS makes the rule unreachable at the positions left out, which is
+ * a silent bug. So an incomplete shape (or none at all) is always correct, and
+ * only an over-narrow one is wrong. `NormalizerIndexFidelity.spec` checks every
+ * declaration against its rule's real behaviour.
+ */
+interface RuleSlot {
+    /**
+     * Token types admitted at this slot. Omit when the type is unconstrained,
+     * or when the rule accepts so many that naming them filters nothing.
+     */
+    readonly types?: readonly string[];
+    /**
+     * Token values admitted at this slot, compared case-insensitively.
+     *
+     * This is the axis that separates rules sharing a start type. The
+     * call-fusion rules all begin at an `IDENT`, the commonest token in prose,
+     * so type alone leaves every one of them a candidate at every word; the word
+     * itself is what tells them apart, and each already owns that set as an
+     * exported constant.
+     *
+     * A rule whose own check is case-SENSITIVE (the stock ticker rule matches
+     * upper case only) may still declare its lower-cased words here: the index
+     * only ever over-approximates, and the rule's own check still runs.
+     */
+    readonly values?: readonly string[];
+}
 /**
  * A pluggable normalization rule registered with the TokenNormalizer.
  *
@@ -139,6 +176,60 @@ interface NormalizerRule {
      * the rule silently unreachable there, which is a bug.
      */
     readonly startTokenTypes?: readonly string[];
+    /**
+     * The rule's leading shape: what the tokens from the match position onward
+     * may be, one {@link RuleSlot} per position.
+     *
+     * This generalises {@link startTokenTypes}, which constrains only the first
+     * token. Constraining the first token alone is not enough to separate the
+     * rules that matter: every rule firing on a bare `NUMBER` declares the same
+     * start type, so they all remain candidates at every number in the document.
+     * What distinguishes them is the token after it, `NUMBER COLON` being a clock
+     * time and `NUMBER SLASH` a network address, and that fact is only usable by
+     * an index if the rule states it rather than hiding it inside `match()`.
+     *
+     * Depth is the rule's choice, not the interface's. The normalizer builds one
+     * lookup plane per declared slot and intersects them, so a rule that declares
+     * three positions is filtered on three. It may also index fewer planes than
+     * were declared, which stays correct for the reason given on {@link RuleSlot}:
+     * a shallower filter admits more candidates, and each surviving rule still
+     * runs its own `match()`.
+     *
+     * Prefer this to {@link startTokenTypes} in new rules. When both are given,
+     * this wins; `startTokenTypes: ["IDENT"]` means exactly `shape: [{ types:
+     * ["IDENT"] }]`.
+     *
+     * @example
+     * ```ts
+     * // 9:00am, 16:00, a clock time is a number followed by a colon
+     * shape: [{ types: ["NUMBER"] }, { types: ["COLON"] }]
+     *
+     * // sha256("hi"), a known word followed by an opening parenthesis
+     * shape: [{ types: ["IDENT"], values: HASH_NAMES }, { types: ["LPAREN"] }]
+     * ```
+     */
+    readonly shape?: readonly RuleSlot[];
+    /**
+     * Why this rule cannot declare a {@link shape}, for the few that genuinely
+     * cannot.
+     *
+     * A rule with neither a shape nor a `startTokenTypes` hint is tried at every
+     * position of every line, so it raises the cost of the whole document rather
+     * than only its own feature. Registering one logs a warning naming the rule,
+     * which is how a package author finds out before their users do.
+     *
+     * Some rules really cannot be described by a leading shape: an unbounded
+     * forward scan, a greedy match against a table the host mutates at runtime.
+     * Setting this states that case, silences the warning, and leaves the reason
+     * in the code where the next person will read it. It is deliberately a
+     * sentence and not a boolean, because "why" is the part worth keeping.
+     *
+     * @example
+     * ```ts
+     * unshapedReason: "Scans forward an unbounded number of NUMBER UNIT pairs, so no fixed leading shape describes it.",
+     * ```
+     */
+    readonly unshapedReason?: string;
     /**
      * Attempt to match a pattern starting at position `pos` in the token stream.
      *
@@ -228,6 +319,19 @@ interface NormalizerOptions {
      * When `undefined`, fusions are still tracked internally but no callbacks fire.
      */
     onFusion?: (fusion: TokenFusion) => void;
+    /**
+     * Try every registered rule at every position, ignoring the shape index.
+     *
+     * Diagnostic only, and much slower. It exists so the indexed walk can be
+     * compared against the unindexed one over a corpus: the index is a pure
+     * filter, so the two must agree token for token, and a rule whose declared
+     * {@link NormalizerRule.shape} is too narrow shows up as a difference rather
+     * than as a feature that quietly stopped working.
+     * `NormalizerIndexFidelity.spec` is the consumer.
+     *
+     * @default false
+     */
+    ignoreRuleIndex?: boolean;
 }
 /**
  * Creates a new normalized token from fused source tokens.
@@ -298,6 +402,28 @@ declare class TokenNormalizer {
      * identifier. Invalidated alongside {@link sortedRulesCache}.
      */
     private rulesByTokenType;
+    /**
+     * Shape index over the priority-sorted rules, rebuilt alongside
+     * {@link sortedRulesCache}. `null` means "stale, rebuild on next use".
+     *
+     * This is what turns the per-position scan from "try every rule that could
+     * fire on this token type" into "AND a few lookup planes and try what
+     * survives", which is usually nothing. See {@link RuleIndex}.
+     */
+    private ruleIndexCache;
+    /** Reused by {@link rulesAt} so a position's candidate list costs no allocation. */
+    private candidateBuffer;
+    /**
+     * How many times {@link normalize} has run, capped once the index is in use.
+     *
+     * The index costs a few tens of microseconds to build and pays that back
+     * within a line or two, but an engine that normalises exactly once, which is
+     * what evaluating a single expression on a fresh engine does, would never
+     * reach the payback. Skipping the build on the first call keeps that case at
+     * the cost it had before the index existed, and a document reaches line two
+     * immediately.
+     */
+    private normalizeCalls;
     /**
      * Phrase trie for single-pass multi-word phrase fusion.
      * Tried at each token position BEFORE other rules, the trie walk
@@ -348,11 +474,67 @@ declare class TokenNormalizer {
      * and cached, which is what turns the per-position rule scan from "try all R
      * rules" into "try only the ones that could fire on this token".
      */
+    private getRuleIndex;
+    /**
+     * The rules to try at a position, in priority order.
+     *
+     * With a candidate mask, walks its set bits low to high, which is descending
+     * priority because bit `i` is the rule at index `i` of the priority-sorted
+     * list.
+     *
+     * The walk shifts a bit at a time rather than jumping to the next set bit
+     * with `Math.clz32(bits & -bits)`, which is the idiomatic form and was 36x
+     * SLOWER here: 45us per line against 1.2us, measured over the built-in rule
+     * set. `clz32` is specified on uint32 and the masks come out of a
+     * `Uint32Array` as doubles above 2^31, so every call pays a conversion that
+     * swamps the handful of iterations it saves. A de Bruijn table matched the
+     * shift scan to within noise and needs a magic constant, so the plain shift
+     * wins on both counts. Worst case is 32 iterations per word of pure integer
+     * work.
+     *
+     * With `null` (the {@link NormalizerOptions.ignoreRuleIndex} path) it falls
+     * back to the type-bucketed list, which is the behaviour this replaced.
+     *
+     * Returns a buffer reused across positions, so the caller must finish with it
+     * before calling again. Copying the surviving rules out here rather than
+     * yielding them lazily is deliberate twice over: it keeps the mask's own
+     * scratch buffer from being read after the next position overwrites it, and
+     * it avoids a generator on the hottest loop in the pass.
+     */
+    private rulesAt;
     private rulesForTokenType;
     /**
      * Get the number of currently registered rules (excludes phrase trie entries).
      */
     get ruleCount(): number;
+    /**
+     * Every registered rule with the shape it declared, for diagnostic display.
+     *
+     * Exposes what the index is actually working with: a rule that declares a
+     * shape is only tried where that shape can match, and one that declares none
+     * is tried at every position of every line. Which is which is invisible from
+     * the outside otherwise, and it is the difference between a package that
+     * costs the documents that use it and one that costs all of them, so the
+     * playground draws it.
+     *
+     * Returned in priority order, highest first, which is the order the
+     * normalizer tries them in.
+     */
+    private ruleShapesCache;
+    getRuleShapes(): Array<{
+        name: string;
+        priority: number;
+        shape: readonly RuleSlot[];
+        unshapedReason?: string;
+        indexedSlots: number;
+    }>;
+    /**
+     * How many rules could fire at each position of `tokens`, against the total.
+     *
+     * The point of the index is that most positions admit no rule at all, and
+     * this is what makes that visible rather than asserted.
+     */
+    getCandidateCounts(tokens: Token[]): number[];
     /**
      * Register a multi-word phrase for fusion into a single compound token.
      *

package/dist/{TokenNormalizer-d7F1KQFs.d.ts → TokenNormalizer-OTPS0Otq.d.ts}

@@ -69,6 +69,43 @@ interface NormalizerMatch {
      */
     ruleName?: string;
 }
+/**
+ * One position of a rule's leading shape, as a declarative constraint.
+ *
+ * A slot says what the token at that offset from the match position may be.
+ * Both fields are optional and an omitted one constrains nothing, so `{}` is a
+ * wildcard slot, useful for reaching past a position a rule does not care about
+ * to one it does.
+ *
+ * The exactness contract runs in one direction only, and it is the whole reason
+ * this is safe to adopt gradually. Declaring MORE than the rule can match costs
+ * a `match()` call that returns null, which is what happens today anyway.
+ * Declaring LESS makes the rule unreachable at the positions left out, which is
+ * a silent bug. So an incomplete shape (or none at all) is always correct, and
+ * only an over-narrow one is wrong. `NormalizerIndexFidelity.spec` checks every
+ * declaration against its rule's real behaviour.
+ */
+interface RuleSlot {
+    /**
+     * Token types admitted at this slot. Omit when the type is unconstrained,
+     * or when the rule accepts so many that naming them filters nothing.
+     */
+    readonly types?: readonly string[];
+    /**
+     * Token values admitted at this slot, compared case-insensitively.
+     *
+     * This is the axis that separates rules sharing a start type. The
+     * call-fusion rules all begin at an `IDENT`, the commonest token in prose,
+     * so type alone leaves every one of them a candidate at every word; the word
+     * itself is what tells them apart, and each already owns that set as an
+     * exported constant.
+     *
+     * A rule whose own check is case-SENSITIVE (the stock ticker rule matches
+     * upper case only) may still declare its lower-cased words here: the index
+     * only ever over-approximates, and the rule's own check still runs.
+     */
+    readonly values?: readonly string[];
+}
 /**
  * A pluggable normalization rule registered with the TokenNormalizer.
  *
@@ -139,6 +176,60 @@ interface NormalizerRule {
      * the rule silently unreachable there, which is a bug.
      */
     readonly startTokenTypes?: readonly string[];
+    /**
+     * The rule's leading shape: what the tokens from the match position onward
+     * may be, one {@link RuleSlot} per position.
+     *
+     * This generalises {@link startTokenTypes}, which constrains only the first
+     * token. Constraining the first token alone is not enough to separate the
+     * rules that matter: every rule firing on a bare `NUMBER` declares the same
+     * start type, so they all remain candidates at every number in the document.
+     * What distinguishes them is the token after it, `NUMBER COLON` being a clock
+     * time and `NUMBER SLASH` a network address, and that fact is only usable by
+     * an index if the rule states it rather than hiding it inside `match()`.
+     *
+     * Depth is the rule's choice, not the interface's. The normalizer builds one
+     * lookup plane per declared slot and intersects them, so a rule that declares
+     * three positions is filtered on three. It may also index fewer planes than
+     * were declared, which stays correct for the reason given on {@link RuleSlot}:
+     * a shallower filter admits more candidates, and each surviving rule still
+     * runs its own `match()`.
+     *
+     * Prefer this to {@link startTokenTypes} in new rules. When both are given,
+     * this wins; `startTokenTypes: ["IDENT"]` means exactly `shape: [{ types:
+     * ["IDENT"] }]`.
+     *
+     * @example
+     * ```ts
+     * // 9:00am, 16:00, a clock time is a number followed by a colon
+     * shape: [{ types: ["NUMBER"] }, { types: ["COLON"] }]
+     *
+     * // sha256("hi"), a known word followed by an opening parenthesis
+     * shape: [{ types: ["IDENT"], values: HASH_NAMES }, { types: ["LPAREN"] }]
+     * ```
+     */
+    readonly shape?: readonly RuleSlot[];
+    /**
+     * Why this rule cannot declare a {@link shape}, for the few that genuinely
+     * cannot.
+     *
+     * A rule with neither a shape nor a `startTokenTypes` hint is tried at every
+     * position of every line, so it raises the cost of the whole document rather
+     * than only its own feature. Registering one logs a warning naming the rule,
+     * which is how a package author finds out before their users do.
+     *
+     * Some rules really cannot be described by a leading shape: an unbounded
+     * forward scan, a greedy match against a table the host mutates at runtime.
+     * Setting this states that case, silences the warning, and leaves the reason
+     * in the code where the next person will read it. It is deliberately a
+     * sentence and not a boolean, because "why" is the part worth keeping.
+     *
+     * @example
+     * ```ts
+     * unshapedReason: "Scans forward an unbounded number of NUMBER UNIT pairs, so no fixed leading shape describes it.",
+     * ```
+     */
+    readonly unshapedReason?: string;
     /**
      * Attempt to match a pattern starting at position `pos` in the token stream.
      *
@@ -228,6 +319,19 @@ interface NormalizerOptions {
      * When `undefined`, fusions are still tracked internally but no callbacks fire.
      */
     onFusion?: (fusion: TokenFusion) => void;
+    /**
+     * Try every registered rule at every position, ignoring the shape index.
+     *
+     * Diagnostic only, and much slower. It exists so the indexed walk can be
+     * compared against the unindexed one over a corpus: the index is a pure
+     * filter, so the two must agree token for token, and a rule whose declared
+     * {@link NormalizerRule.shape} is too narrow shows up as a difference rather
+     * than as a feature that quietly stopped working.
+     * `NormalizerIndexFidelity.spec` is the consumer.
+     *
+     * @default false
+     */
+    ignoreRuleIndex?: boolean;
 }
 /**
  * Creates a new normalized token from fused source tokens.
@@ -298,6 +402,28 @@ declare class TokenNormalizer {
      * identifier. Invalidated alongside {@link sortedRulesCache}.
      */
     private rulesByTokenType;
+    /**
+     * Shape index over the priority-sorted rules, rebuilt alongside
+     * {@link sortedRulesCache}. `null` means "stale, rebuild on next use".
+     *
+     * This is what turns the per-position scan from "try every rule that could
+     * fire on this token type" into "AND a few lookup planes and try what
+     * survives", which is usually nothing. See {@link RuleIndex}.
+     */
+    private ruleIndexCache;
+    /** Reused by {@link rulesAt} so a position's candidate list costs no allocation. */
+    private candidateBuffer;
+    /**
+     * How many times {@link normalize} has run, capped once the index is in use.
+     *
+     * The index costs a few tens of microseconds to build and pays that back
+     * within a line or two, but an engine that normalises exactly once, which is
+     * what evaluating a single expression on a fresh engine does, would never
+     * reach the payback. Skipping the build on the first call keeps that case at
+     * the cost it had before the index existed, and a document reaches line two
+     * immediately.
+     */
+    private normalizeCalls;
     /**
      * Phrase trie for single-pass multi-word phrase fusion.
      * Tried at each token position BEFORE other rules, the trie walk
@@ -348,11 +474,67 @@ declare class TokenNormalizer {
      * and cached, which is what turns the per-position rule scan from "try all R
      * rules" into "try only the ones that could fire on this token".
      */
+    private getRuleIndex;
+    /**
+     * The rules to try at a position, in priority order.
+     *
+     * With a candidate mask, walks its set bits low to high, which is descending
+     * priority because bit `i` is the rule at index `i` of the priority-sorted
+     * list.
+     *
+     * The walk shifts a bit at a time rather than jumping to the next set bit
+     * with `Math.clz32(bits & -bits)`, which is the idiomatic form and was 36x
+     * SLOWER here: 45us per line against 1.2us, measured over the built-in rule
+     * set. `clz32` is specified on uint32 and the masks come out of a
+     * `Uint32Array` as doubles above 2^31, so every call pays a conversion that
+     * swamps the handful of iterations it saves. A de Bruijn table matched the
+     * shift scan to within noise and needs a magic constant, so the plain shift
+     * wins on both counts. Worst case is 32 iterations per word of pure integer
+     * work.
+     *
+     * With `null` (the {@link NormalizerOptions.ignoreRuleIndex} path) it falls
+     * back to the type-bucketed list, which is the behaviour this replaced.
+     *
+     * Returns a buffer reused across positions, so the caller must finish with it
+     * before calling again. Copying the surviving rules out here rather than
+     * yielding them lazily is deliberate twice over: it keeps the mask's own
+     * scratch buffer from being read after the next position overwrites it, and
+     * it avoids a generator on the hottest loop in the pass.
+     */
+    private rulesAt;
     private rulesForTokenType;
     /**
      * Get the number of currently registered rules (excludes phrase trie entries).
      */
     get ruleCount(): number;
+    /**
+     * Every registered rule with the shape it declared, for diagnostic display.
+     *
+     * Exposes what the index is actually working with: a rule that declares a
+     * shape is only tried where that shape can match, and one that declares none
+     * is tried at every position of every line. Which is which is invisible from
+     * the outside otherwise, and it is the difference between a package that
+     * costs the documents that use it and one that costs all of them, so the
+     * playground draws it.
+     *
+     * Returned in priority order, highest first, which is the order the
+     * normalizer tries them in.
+     */
+    private ruleShapesCache;
+    getRuleShapes(): Array<{
+        name: string;
+        priority: number;
+        shape: readonly RuleSlot[];
+        unshapedReason?: string;
+        indexedSlots: number;
+    }>;
+    /**
+     * How many rules could fire at each position of `tokens`, against the total.
+     *
+     * The point of the index is that most positions admit no rule at all, and
+     * this is what makes that visible rather than asserted.
+     */
+    getCandidateCounts(tokens: Token[]): number[];
     /**
      * Register a multi-word phrase for fusion into a single compound token.
      *

package/dist/{VMCheckpoints-DJdvpliw.d.ts → VMCheckpoints-BDRY1Kx8.d.ts}

@@ -1,6 +1,6 @@
 import { V as Value } from './Value-DCTqTSeP.js';
-import { V as VM } from './ScopeManager-C6c1WmJD.js';
-import { U as UserFunctionDef } from './BytecodeBuilder-CRYfrFfq.js';
+import { V as VM } from './ScopeManager-gB9UengK.js';
+import { U as UserFunctionDef } from './BytecodeBuilder-aqVa7Plx.js';
 
 /**
  * A point-in-time snapshot of VM variable state.

package/dist/{VMCheckpoints-BFDlaGse.d.cts → VMCheckpoints-DGar9Yg8.d.cts}

@@ -1,6 +1,6 @@
 import { V as Value } from './Value-DCTqTSeP.cjs';
-import { V as VM } from './ScopeManager-BtqiTVjG.cjs';
-import { U as UserFunctionDef } from './BytecodeBuilder-CRYfrFfq.cjs';
+import { V as VM } from './ScopeManager-bCYewWVt.cjs';
+import { U as UserFunctionDef } from './BytecodeBuilder-aqVa7Plx.cjs';
 
 /**
  * A point-in-time snapshot of VM variable state.

package/dist/{WorkerError-FH7KXX4_.d.cts → WorkerError-DGBNM3gA.d.cts}

@@ -1,4 +1,4 @@
-import { c as ErrorCategory, S as SourceSpan, E as EngineError } from './EngineError-Cv5q4Rbv.cjs';
+import { c as ErrorCategory, S as SourceSpan, E as EngineError } from './EngineError-DTk7I7hZ.cjs';
 
 /**
  * Structured errors for the off-main-thread worker harness, and the pair of

package/dist/{WorkerError-J3v3ix_z.d.ts → WorkerError-DpZgKRks.d.ts}

@@ -1,4 +1,4 @@
-import { c as ErrorCategory, S as SourceSpan, E as EngineError } from './EngineError-Cv5q4Rbv.js';
+import { c as ErrorCategory, S as SourceSpan, E as EngineError } from './EngineError-DTk7I7hZ.js';
 
 /**
  * Structured errors for the off-main-thread worker harness, and the pair of

package/dist/{chunk-UZSPBFZN.js → chunk-2DPKJ2SM.js}

@@ -1,2 +1,2 @@
-import {d}from'./chunk-JPFCXLJV.js';import {b}from'./chunk-TVE2DNPU.js';import {a}from'./chunk-LHLW7JLS.js';import {a as a$1}from'./chunk-UWAXJX5G.js';var u=(s=>(s.Main="main",s.Inline="inline",s.String="string",s))(u||{});var l=class{constructor(e="en",t){this.currentState="main";this.hasPeeked=false;this.tokens=[];this.tokenIdx=0;this.expressionLexer=new b(e,t);}reset(e,t){let s=t??"main";if(this.currentState=s,this.hasPeeked=false,this.peekedToken=void 0,s==="main"){if(this.expressionLexer.classifyLine(e).skip){this.tokens=[],this.tokenIdx=0;return}this.expressionLexer.reset(e),this.tokens=this.expressionLexer.tokenizeAll(),this.tokenIdx=0;}else this.expressionLexer.reset(e),this.tokens=this.expressionLexer.tokenizeAll(),this.tokenIdx=0;}classifyLine(e){return this.expressionLexer.classifyLine(e)}findInlineSolves(e){return this.expressionLexer.findInlineSolves(e)}getKeywords(){return this.expressionLexer.getKeywords()}next(){if(this.hasPeeked)return this.hasPeeked=false,this.peekedToken;if(this.tokenIdx<this.tokens.length)return this.tokens[this.tokenIdx++]}peek(){return this.hasPeeked?this.peekedToken:(this.peekedToken=this.next(),this.hasPeeked=true,this.peekedToken)}[Symbol.iterator](){return this.tokens[Symbol.iterator]()}registerVocabulary(e){this.expressionLexer.registerVocabulary(e);}unregisterVocabulary(e){this.expressionLexer.unregisterVocabulary(e);}resetExpression(e){this.currentState="main",this.hasPeeked=false,this.peekedToken=void 0,this.expressionLexer.reset(e),this.tokens=this.expressionLexer.tokenizeAll(),this.tokenIdx=0;}scanDocument(e){return this.expressionLexer.scanDocument(e)}getState(){return this.currentState}setState(e){this.currentState=e;}getHighlightTokens(e){let t=this.expressionLexer.classifyLine(e);return t.skip&&e.startsWith("> ")?this.collectHighlightTokens(e.slice(2)):t.skip?[]:this.collectHighlightTokens(e)}getHighlightTokenObjects(e){let t=this.expressionLexer.classifyLine(e);return t.skip&&e.startsWith("> ")?this.collectTokenObjects(e.slice(2)):t.skip?[]:this.collectTokenObjects(e)}collectTokenObjects(e){this.resetExpression(e);let t=[];for(let s of this)s.type==="WS"||s.type==="NEWLINE"||s.type.startsWith("MD_")||s.type==="INLINE_SOLVE_START"||s.type==="BACKTICK_CLOSE"||t.push(s);return t}collectHighlightTokens(e){return this.collectTokenObjects(e).map(t=>({type:t.type,value:t.value,offset:t.offset,col:t.col,length:t.text.length,category:d(t.type)}))}},T=new l("en",void 0);var n=class{constructor(){this.classes=[];this.localeKeywordMap=null;this.localePhraseMap=null;this.unitNames=null;}register(e){this.classes.push(e);}unregister(e){this.classes=this.classes.filter(t=>t.tokenType!==e);}setLocale(e,t){this.localeKeywordMap=e,this.localePhraseMap=t??null;}setUnits(e){this.unitNames=e;}build(){let e=new Map;if(this.localeKeywordMap)for(let[r,i]of Object.entries(this.localeKeywordMap))e.set(r.toLowerCase(),i);let t=[...this.classes].sort((r,i)=>(r.priority??0)-(i.priority??0));for(let r of t)for(let i of Object.keys(r.keywords))e.set(i.toLowerCase(),r.tokenType);let s=this.buildPhraseTrie(),o=this.buildPhraseStartWords();return {keywordToType:e,phraseTrie:s,phraseStartWords:o,unitNames:this.unitNames??new Set}}buildPhraseTrie(){let e={children:new Map};if(this.localePhraseMap)for(let[t,s]of Object.entries(this.localePhraseMap))this.insertPhrase(e,t.toLowerCase(),s);for(let t of this.classes)if(t.phrases)for(let s of Object.keys(t.phrases))this.insertPhrase(e,s.toLowerCase(),t.tokenType);return e.children.size>0?e:null}insertPhrase(e,t,s){let o=t.split(" "),r=e;for(let i of o)r.children.has(i)||r.children.set(i,{children:new Map}),r=r.children.get(i);r.type||(r.type=s);}buildPhraseStartWords(){let e=new Set;if(this.localePhraseMap)for(let t of Object.keys(this.localePhraseMap)){let s=t.split(" ")[0].toLowerCase();e.add(s);}for(let t of this.classes)if(t.phrases)for(let s of Object.keys(t.phrases)){let o=s.split(" ")[0].toLowerCase();e.add(o);}return e}};var k={"to the power of":"CARET","power of":"CARET","increase by":"INCREASE_BY","decrease by":"DECREASE_BY","times by":"TIMES_BY","multiply by":"MULTIPLY_BY","multiplied by":"MULTIPLY_BY","divide by":"DIVIDE_BY"};function P(a$2="en"){let e=a(a$2),t=new n;return t.setLocale(e.keywordMap,k),t.setUnits(a$1),t.build()}export{u as a,l as b,T as c,P as d};//# sourceMappingURL=chunk-UZSPBFZN.js.map
-//# sourceMappingURL=chunk-UZSPBFZN.js.map
+import {d}from'./chunk-JPFCXLJV.js';import {b}from'./chunk-LJFS3XHW.js';import {a}from'./chunk-LHLW7JLS.js';import {a as a$1}from'./chunk-UWAXJX5G.js';var u=(s=>(s.Main="main",s.Inline="inline",s.String="string",s))(u||{});var l=class{constructor(e="en",t){this.currentState="main";this.hasPeeked=false;this.tokens=[];this.tokenIdx=0;this.expressionLexer=new b(e,t);}reset(e,t){let s=t??"main";if(this.currentState=s,this.hasPeeked=false,this.peekedToken=void 0,s==="main"){if(this.expressionLexer.classifyLine(e).skip){this.tokens=[],this.tokenIdx=0;return}this.expressionLexer.reset(e),this.tokens=this.expressionLexer.tokenizeAll(),this.tokenIdx=0;}else this.expressionLexer.reset(e),this.tokens=this.expressionLexer.tokenizeAll(),this.tokenIdx=0;}classifyLine(e){return this.expressionLexer.classifyLine(e)}findInlineSolves(e){return this.expressionLexer.findInlineSolves(e)}getKeywords(){return this.expressionLexer.getKeywords()}next(){if(this.hasPeeked)return this.hasPeeked=false,this.peekedToken;if(this.tokenIdx<this.tokens.length)return this.tokens[this.tokenIdx++]}peek(){return this.hasPeeked?this.peekedToken:(this.peekedToken=this.next(),this.hasPeeked=true,this.peekedToken)}[Symbol.iterator](){return this.tokens[Symbol.iterator]()}registerVocabulary(e){this.expressionLexer.registerVocabulary(e);}unregisterVocabulary(e){this.expressionLexer.unregisterVocabulary(e);}resetExpression(e){this.currentState="main",this.hasPeeked=false,this.peekedToken=void 0,this.expressionLexer.reset(e),this.tokens=this.expressionLexer.tokenizeAll(),this.tokenIdx=0;}scanDocument(e){return this.expressionLexer.scanDocument(e)}getState(){return this.currentState}setState(e){this.currentState=e;}getHighlightTokens(e){let t=this.expressionLexer.classifyLine(e);return t.skip&&e.startsWith("> ")?this.collectHighlightTokens(e.slice(2)):t.skip?[]:this.collectHighlightTokens(e)}getHighlightTokenObjects(e){let t=this.expressionLexer.classifyLine(e);return t.skip&&e.startsWith("> ")?this.collectTokenObjects(e.slice(2)):t.skip?[]:this.collectTokenObjects(e)}collectTokenObjects(e){this.resetExpression(e);let t=[];for(let s of this)s.type==="WS"||s.type==="NEWLINE"||s.type.startsWith("MD_")||s.type==="INLINE_SOLVE_START"||s.type==="BACKTICK_CLOSE"||t.push(s);return t}collectHighlightTokens(e){return this.collectTokenObjects(e).map(t=>({type:t.type,value:t.value,offset:t.offset,col:t.col,length:t.text.length,category:d(t.type)}))}},T=new l("en",void 0);var n=class{constructor(){this.classes=[];this.localeKeywordMap=null;this.localePhraseMap=null;this.unitNames=null;}register(e){this.classes.push(e);}unregister(e){this.classes=this.classes.filter(t=>t.tokenType!==e);}setLocale(e,t){this.localeKeywordMap=e,this.localePhraseMap=t??null;}setUnits(e){this.unitNames=e;}build(){let e=new Map;if(this.localeKeywordMap)for(let[r,i]of Object.entries(this.localeKeywordMap))e.set(r.toLowerCase(),i);let t=[...this.classes].sort((r,i)=>(r.priority??0)-(i.priority??0));for(let r of t)for(let i of Object.keys(r.keywords))e.set(i.toLowerCase(),r.tokenType);let s=this.buildPhraseTrie(),o=this.buildPhraseStartWords();return {keywordToType:e,phraseTrie:s,phraseStartWords:o,unitNames:this.unitNames??new Set}}buildPhraseTrie(){let e={children:new Map};if(this.localePhraseMap)for(let[t,s]of Object.entries(this.localePhraseMap))this.insertPhrase(e,t.toLowerCase(),s);for(let t of this.classes)if(t.phrases)for(let s of Object.keys(t.phrases))this.insertPhrase(e,s.toLowerCase(),t.tokenType);return e.children.size>0?e:null}insertPhrase(e,t,s){let o=t.split(" "),r=e;for(let i of o)r.children.has(i)||r.children.set(i,{children:new Map}),r=r.children.get(i);r.type||(r.type=s);}buildPhraseStartWords(){let e=new Set;if(this.localePhraseMap)for(let t of Object.keys(this.localePhraseMap)){let s=t.split(" ")[0].toLowerCase();e.add(s);}for(let t of this.classes)if(t.phrases)for(let s of Object.keys(t.phrases)){let o=s.split(" ")[0].toLowerCase();e.add(o);}return e}};var k={"to the power of":"CARET","power of":"CARET","increase by":"INCREASE_BY","decrease by":"DECREASE_BY","times by":"TIMES_BY","multiply by":"MULTIPLY_BY","multiplied by":"MULTIPLY_BY","divide by":"DIVIDE_BY"};function P(a$2="en"){let e=a(a$2),t=new n;return t.setLocale(e.keywordMap,k),t.setUnits(a$1),t.build()}export{u as a,l as b,T as c,P as d};//# sourceMappingURL=chunk-2DPKJ2SM.js.map
+//# sourceMappingURL=chunk-2DPKJ2SM.js.map

package/dist/{chunk-UZSPBFZN.js.map → chunk-2DPKJ2SM.js.map}

@@ -1 +1 @@
-{"version":3,"sources":["../src/lexer/LexerState.ts","../src/lexer/Lexer.ts","../src/lexer/TokenClassRegistry.ts","../src/lexer/tokenRegistration.ts"],"names":["LexerState","Lexer","localeCode","tokenLookup","ExpressionLexer","input","state","newState","lineText","plugin","text","classification","result","token","getTokenCategory","sharedLexer","TokenClassRegistry","tokenClass","tokenType","c","keywordMap","phraseMap","unitNames","keywordToType","keyword","sorted","a","b","tc","phraseTrie","phraseStartWords","root","phrase","words","node","word","startWords","first","BUILTIN_PHRASES","buildTokenLookup","locale","getLocale","registry","knownUnits"],"mappings":"uJAMO,IAAKA,CAAAA,CAAAA,CAAAA,CAAAA,GACXA,EAAA,IAAA,CAAO,MAAA,CACPA,EAAA,MAAA,CAAS,QAAA,CACTA,EAAA,MAAA,CAAS,QAAA,CAHEA,OAAA,EAAA,ECaL,IAAMC,EAAN,KAAY,CAkBjB,YAAYC,CAAAA,CAAa,IAAA,CAAMC,EAA2B,CAf1D,IAAA,CAAQ,aAA2B,MAAA,CAEnC,IAAA,CAAQ,UAAY,KAAA,CAIpB,IAAA,CAAQ,OAAkB,EAAC,CAC3B,KAAQ,QAAA,CAAmB,CAAA,CAYzB,KAAK,eAAA,CAAkB,IAAIC,EAAgBF,CAAAA,CAAYC,CAAW,EACpE,CAEA,KAAA,CAAME,EAAeC,CAAAA,CAA0B,CAC7C,IAAMC,CAAAA,CAAWD,CAAAA,EAAS,OAQ1B,GAPA,IAAA,CAAK,aAAeC,CAAAA,CACpB,IAAA,CAAK,UAAY,KAAA,CACjB,IAAA,CAAK,YAAc,MAAA,CAKfA,CAAAA,GAAa,OAAiB,CAEhC,GADuB,KAAK,eAAA,CAAgB,YAAA,CAAaF,CAAK,CAAA,CAC3C,IAAA,CAAM,CACvB,IAAA,CAAK,MAAA,CAAS,EAAC,CACf,IAAA,CAAK,SAAW,CAAA,CAChB,MACF,CAEA,IAAA,CAAK,eAAA,CAAgB,MAAMA,CAAK,CAAA,CAChC,KAAK,MAAA,CAAS,IAAA,CAAK,gBAAgB,WAAA,EAAY,CAC/C,KAAK,QAAA,CAAW,EAClB,MAEE,IAAA,CAAK,eAAA,CAAgB,MAAMA,CAAK,CAAA,CAChC,KAAK,MAAA,CAAS,IAAA,CAAK,gBAAgB,WAAA,EAAY,CAC/C,KAAK,QAAA,CAAW,EAEpB,CAMA,YAAA,CAAaG,CAAAA,CAAsC,CACjD,OAAO,IAAA,CAAK,gBAAgB,YAAA,CAAaA,CAAQ,CACnD,CAMA,gBAAA,CAAiBA,EAAkB,CACjC,OAAO,KAAK,eAAA,CAAgB,gBAAA,CAAiBA,CAAQ,CACvD,CAMA,aAAsC,CACpC,OAAO,KAAK,eAAA,CAAgB,WAAA,EAC9B,CAEA,IAAA,EAA0B,CACxB,GAAI,IAAA,CAAK,UACP,OAAA,IAAA,CAAK,SAAA,CAAY,MACV,IAAA,CAAK,WAAA,CAGd,GAAI,IAAA,CAAK,QAAA,CAAW,KAAK,MAAA,CAAO,MAAA,CAC9B,OAAO,IAAA,CAAK,MAAA,CAAO,KAAK,QAAA,EAAU,CAGtC,CAEA,IAAA,EAA0B,CACxB,OAAI,IAAA,CAAK,SAAA,CAAkB,KAAK,WAAA,EAChC,IAAA,CAAK,YAAc,IAAA,CAAK,IAAA,GACxB,IAAA,CAAK,SAAA,CAAY,KACV,IAAA,CAAK,WAAA,CACd,CAEA,CAAC,MAAA,CAAO,QAAQ,CAAA,EAAqB,CACnC,OAAO,IAAA,CAAK,MAAA,CAAO,MAAA,CAAO,QAAQ,CAAA,EACpC,CAQA,kBAAA,CAAmBC,CAAAA,CAA+B,CAChD,IAAA,CAAK,eAAA,CAAgB,mBAAmBA,CAAM,EAChD,CAMA,oBAAA,CAAqBA,CAAAA,CAA+B,CAClD,IAAA,CAAK,eAAA,CAAgB,qBAAqBA,CAAM,EAClD,CAOA,eAAA,CAAgBJ,CAAAA,CAAqB,CACnC,IAAA,CAAK,YAAA,CAAe,OACpB,IAAA,CAAK,SAAA,CAAY,MACjB,IAAA,CAAK,WAAA,CAAc,OACnB,IAAA,CAAK,eAAA,CAAgB,MAAMA,CAAK,CAAA,CAChC,KAAK,MAAA,CAAS,IAAA,CAAK,gBAAgB,WAAA,EAAY,CAC/C,KAAK,QAAA,CAAW,EAClB,CAQA,YAAA,CAAaK,CAAAA,CAAgC,CAC3C,OAAO,IAAA,CAAK,gBAAgB,YAAA,CAAaA,CAAI,CAC/C,CAEA,QAAA,EAAuB,CACrB,OAAO,IAAA,CAAK,YACd,CAEA,QAAA,CAASJ,EAAyB,CAChC,IAAA,CAAK,aAAeA,EACtB,CAEA,mBAAmBE,CAAAA,CAAqI,CACtJ,IAAMG,CAAAA,CAAiB,IAAA,CAAK,gBAAgB,YAAA,CAAaH,CAAQ,EAKjE,OAAIG,CAAAA,CAAe,MAAQH,CAAAA,CAAS,UAAA,CAAW,IAAI,CAAA,CAC1C,IAAA,CAAK,uBAAuBA,CAAAA,CAAS,KAAA,CAAM,CAAC,CAAC,CAAA,CAGlDG,EAAe,IAAA,CACV,GAGF,IAAA,CAAK,sBAAA,CAAuBH,CAAQ,CAC7C,CAYA,yBAAyBA,CAAAA,CAA2B,CAClD,IAAMG,CAAAA,CAAiB,IAAA,CAAK,gBAAgB,YAAA,CAAaH,CAAQ,EACjE,OAAIG,CAAAA,CAAe,MAAQH,CAAAA,CAAS,UAAA,CAAW,IAAI,CAAA,CAC1C,IAAA,CAAK,oBAAoBA,CAAAA,CAAS,KAAA,CAAM,CAAC,CAAC,CAAA,CAE/CG,EAAe,IAAA,CAAa,GACzB,IAAA,CAAK,mBAAA,CAAoBH,CAAQ,CAC1C,CAEQ,oBAAoBA,CAAAA,CAA2B,CACrD,KAAK,eAAA,CAAgBA,CAAQ,EAC7B,IAAMI,CAAAA,CAAkB,EAAC,CACzB,IAAA,IAAWC,KAAS,IAAA,CACdA,CAAAA,CAAM,OAAS,IAAA,EAAQA,CAAAA,CAAM,OAAS,SAAA,EACtCA,CAAAA,CAAM,KAAK,UAAA,CAAW,KAAK,GAC3BA,CAAAA,CAAM,IAAA,GAAS,sBAAwBA,CAAAA,CAAM,IAAA,GAAS,kBAC1DD,CAAAA,CAAO,IAAA,CAAKC,CAAK,CAAA,CAEnB,OAAOD,CACT,CAEQ,sBAAA,CAAuBJ,EAAqI,CAClK,OAAO,KAAK,mBAAA,CAAoBA,CAAQ,EAAE,GAAA,CAAIK,CAAAA,GAAU,CACtD,IAAA,CAAMA,CAAAA,CAAM,KACZ,KAAA,CAAOA,CAAAA,CAAM,KAAA,CACb,MAAA,CAAQA,CAAAA,CAAM,MAAA,CACd,IAAKA,CAAAA,CAAM,GAAA,CAIX,OAAQA,CAAAA,CAAM,IAAA,CAAK,OACnB,QAAA,CAAUC,CAAAA,CAAiBD,EAAM,IAAI,CACvC,EAAE,CACJ,CACF,EAcaE,CAAAA,CAAc,IAAId,EAAM,IAAA,CAAM,MAAS,EClJ7C,IAAMe,CAAAA,CAAN,KAAyB,CAAzB,WAAA,EAAA,CACL,KAAQ,OAAA,CAAwB,GAChC,IAAA,CAAQ,gBAAA,CAAkD,KAC1D,IAAA,CAAQ,eAAA,CAAiD,KACzD,IAAA,CAAQ,SAAA,CAAwC,MAUhD,QAAA,CAASC,CAAAA,CAA8B,CACrC,IAAA,CAAK,OAAA,CAAQ,KAAKA,CAAU,EAC9B,CAMA,UAAA,CAAWC,CAAAA,CAAyB,CAClC,IAAA,CAAK,OAAA,CAAU,KAAK,OAAA,CAAQ,MAAA,CAAOC,GAAKA,CAAAA,CAAE,SAAA,GAAcD,CAAS,EACnE,CAMA,UAAUE,CAAAA,CAAoCC,CAAAA,CAA0C,CACtF,IAAA,CAAK,gBAAA,CAAmBD,EACxB,IAAA,CAAK,eAAA,CAAkBC,GAAa,KACtC,CASA,SAASC,CAAAA,CAAsC,CAC7C,KAAK,SAAA,CAAYA,EACnB,CAYA,KAAA,EAAqB,CACnB,IAAMC,CAAAA,CAAgB,IAAI,IAG1B,GAAI,IAAA,CAAK,iBACP,IAAA,GAAW,CAACC,EAASN,CAAS,CAAA,GAAK,OAAO,OAAA,CAAQ,IAAA,CAAK,gBAAgB,CAAA,CACrEK,CAAAA,CAAc,IAAIC,CAAAA,CAAQ,WAAA,GAAeN,CAAS,CAAA,CAKtD,IAAMO,CAAAA,CAAS,CAAC,GAAG,IAAA,CAAK,OAAO,EAAE,IAAA,CAAK,CAACC,EAAGC,CAAAA,GAAAA,CAAOD,CAAAA,CAAE,UAAY,CAAA,GAAMC,CAAAA,CAAE,UAAY,CAAA,CAAE,CAAA,CACrF,QAAWC,CAAAA,IAAMH,CAAAA,CACf,QAAWD,CAAAA,IAAW,MAAA,CAAO,KAAKI,CAAAA,CAAG,QAAQ,EAC3CL,CAAAA,CAAc,GAAA,CAAIC,EAAQ,WAAA,EAAY,CAAGI,EAAG,SAAS,CAAA,CAKzD,IAAMC,CAAAA,CAAa,IAAA,CAAK,iBAAgB,CAGlCC,CAAAA,CAAmB,KAAK,qBAAA,EAAsB,CAEpD,OAAO,CACL,aAAA,CAAAP,EACA,UAAA,CAAAM,CAAAA,CACA,iBAAAC,CAAAA,CACA,SAAA,CAAW,KAAK,SAAA,EAAa,IAAI,GACnC,CACF,CAIQ,iBAAqC,CAC3C,IAAMC,EAAmB,CAAE,QAAA,CAAU,IAAI,GAAM,CAAA,CAG/C,GAAI,IAAA,CAAK,eAAA,CACP,OAAW,CAACC,CAAAA,CAAQd,CAAS,CAAA,GAAK,MAAA,CAAO,QAAQ,IAAA,CAAK,eAAe,CAAA,CACnE,IAAA,CAAK,YAAA,CAAaa,CAAAA,CAAMC,EAAO,WAAA,EAAY,CAAGd,CAAS,CAAA,CAK3D,IAAA,IAAWU,KAAM,IAAA,CAAK,OAAA,CACpB,GAAKA,CAAAA,CAAG,OAAA,CACR,QAAWI,CAAAA,IAAU,MAAA,CAAO,KAAKJ,CAAAA,CAAG,OAAO,EACzC,IAAA,CAAK,YAAA,CAAaG,EAAMC,CAAAA,CAAO,WAAA,GAAeJ,CAAAA,CAAG,SAAS,EAI9D,OAAOG,CAAAA,CAAK,SAAS,IAAA,CAAO,CAAA,CAAIA,EAAO,IACzC,CAEQ,aAAaA,CAAAA,CAAkBC,CAAAA,CAAgBd,EAAyB,CAC9E,IAAMe,EAAQD,CAAAA,CAAO,KAAA,CAAM,GAAG,CAAA,CAC1BE,CAAAA,CAAOH,EACX,IAAA,IAAWI,CAAAA,IAAQF,EACZC,CAAAA,CAAK,QAAA,CAAS,IAAIC,CAAI,CAAA,EACzBD,EAAK,QAAA,CAAS,GAAA,CAAIC,EAAM,CAAE,QAAA,CAAU,IAAI,GAAM,CAAC,EAEjDD,CAAAA,CAAOA,CAAAA,CAAK,SAAS,GAAA,CAAIC,CAAI,EAG1BD,CAAAA,CAAK,IAAA,GACRA,EAAK,IAAA,CAAOhB,CAAAA,EAEhB,CAEQ,qBAAA,EAAqC,CAC3C,IAAMkB,CAAAA,CAAa,IAAI,IAGvB,GAAI,IAAA,CAAK,gBACP,IAAA,IAAWJ,CAAAA,IAAU,OAAO,IAAA,CAAK,IAAA,CAAK,eAAe,CAAA,CAAG,CACtD,IAAMK,CAAAA,CAAQL,CAAAA,CAAO,MAAM,GAAG,CAAA,CAAE,CAAC,CAAA,CAAE,WAAA,GACnCI,CAAAA,CAAW,GAAA,CAAIC,CAAK,EACtB,CAIF,QAAWT,CAAAA,IAAM,IAAA,CAAK,QACpB,GAAKA,CAAAA,CAAG,QACR,IAAA,IAAWI,CAAAA,IAAU,OAAO,IAAA,CAAKJ,CAAAA,CAAG,OAAO,CAAA,CAAG,CAC5C,IAAMS,CAAAA,CAAQL,CAAAA,CAAO,MAAM,GAAG,CAAA,CAAE,CAAC,CAAA,CAAE,WAAA,GACnCI,CAAAA,CAAW,GAAA,CAAIC,CAAK,EACtB,CAGF,OAAOD,CACT,CACF,EClOA,IAAME,CAAAA,CAA0C,CAC9C,iBAAA,CAAmB,OAAA,CACnB,WAAY,OAAA,CACZ,aAAA,CAAe,cACf,aAAA,CAAe,aAAA,CACf,WAAY,UAAA,CACZ,aAAA,CAAe,cACf,eAAA,CAAiB,aAAA,CACjB,YAAa,WACf,CAAA,CAiBO,SAASC,CAAAA,CAAiBrC,GAAAA,CAAa,KAAmB,CAC/D,IAAMsC,EAAkBC,CAAAA,CAAUvC,GAAU,EACtCwC,CAAAA,CAAW,IAAI1B,EAGrB,OAAA0B,CAAAA,CAAS,UAAUF,CAAAA,CAAO,UAAA,CAAYF,CAAe,CAAA,CAGrDI,CAAAA,CAAS,SAASC,GAAU,CAAA,CAGrBD,CAAAA,CAAS,KAAA,EAClB","file":"chunk-UZSPBFZN.js","sourcesContent":["/**\n * Lexer state machine modes.\n * - Main: document-level scanning with markdown classification\n * - Inline: expression embedded in markdown inline solve (`s\\`...\\``)\n * - String: inside a double-quoted string literal\n */\nexport enum LexerState {\n\tMain = \"main\",\n\tInline = \"inline\",\n\tString = \"string\",\n}\n","import { ExpressionLexer, LineClassification, LexerVocabulary, type ScanLineResult } from \"./ExpressionLexer\";\nimport { Token } from \"@solve-js/lexer/Token\";\nimport { LexerState } from \"@solve-js/lexer/LexerState\";\nimport { getTokenCategory } from \"@solve-js/language/TokenCategoryMap\";\nimport type { TokenCategory } from \"@solve-js/language/TokenCategory\";\nimport type { TokenLookup } from \"@solve-js/lexer/TokenClassRegistry\";\n\n/**\n * Public tokenizer wrapper around {@link ExpressionLexer}.\n *\n * `ExpressionLexer` does the actual character-by-character scanning;\n * `Lexer` adds a materialized-token-array streaming interface\n * (`next()`/`peek()`) plus line-classification state (`reset()`) so\n * callers can iterate a line's tokens without re-scanning on each peek.\n *\n * Each `ExpressionEngine` instance owns its own `Lexer`, and packages\n * extend it via {@link registerVocabulary} (keywords, operators, units)\n * see `IEnginePackage.lexerVocabulary`.\n */\nexport class Lexer {\n  /** Expression-mode lexer (Phase A: V8-optimized, replaces moo) */\n  private expressionLexer: ExpressionLexer;\n  private currentState: LexerState = LexerState.Main;\n  private peekedToken: Token | undefined;\n  private hasPeeked = false;\n\n  // Materialized token array from the last reset() call, used for\n  // next()/peek() streaming access.\n  private tokens: Token[] = [];\n  private tokenIdx: number = 0;\n\n  /**\n   * @param localeCode - Locale code (e.g., \"en\", \"de\"). Defaults to \"en\".\n   * @param tokenLookup - Optional TokenLookup from TokenClassRegistry.\n   *   When provided, configures ExpressionLexer to use registry-built\n   *   keyword/unit/phrase lookups instead of internal instance maps.\n   */\n  constructor(localeCode = \"en\", tokenLookup?: TokenLookup) {\n    // Pass the lookup directly to ExpressionLexer's constructor, it's an\n    // instance field now, not a static. Each Lexer instance gets its own\n    // isolated lookup, preventing cross-instance corruption.\n    this.expressionLexer = new ExpressionLexer(localeCode, tokenLookup);\n  }\n\n  reset(input: string, state?: LexerState): void {\n    const newState = state ?? LexerState.Main;\n    this.currentState = newState;\n    this.hasPeeked = false;\n    this.peekedToken = undefined;\n\n    // Phase B: Main state classifies the line with the markdown scanner.\n    // Skip lines (headings, fences, HRs, etc.) produce empty token arrays.\n    // Expression lines and lines with inline solves are tokenized normally.\n    if (newState === LexerState.Main) {\n      const classification = this.expressionLexer.classifyLine(input);\n      if (classification.skip) {\n        this.tokens = [];\n        this.tokenIdx = 0;\n        return;\n      }\n      // Expression line or markdown line with inline solves, tokenize.\n      this.expressionLexer.reset(input);\n      this.tokens = this.expressionLexer.tokenizeAll();\n      this.tokenIdx = 0;\n    } else {\n      // Non-main states (Inline, String), expression tokenization.\n      this.expressionLexer.reset(input);\n      this.tokens = this.expressionLexer.tokenizeAll();\n      this.tokenIdx = 0;\n    }\n  }\n\n  /**\n   * Classify a single line of markdown text (Phase B).\n   * Delegates to the ExpressionLexer's character-by-character scanner.\n   */\n  classifyLine(lineText: string): LineClassification {\n    return this.expressionLexer.classifyLine(lineText);\n  }\n\n  /**\n   * Find all inline solve markers in a line (Phase B).\n   * Delegates to the ExpressionLexer's character-by-character scanner.\n   */\n  findInlineSolves(lineText: string) {\n    return this.expressionLexer.findInlineSolves(lineText);\n  }\n\n  /**\n   * Every keyword this lexer currently recognizes (locale + plugin-contributed),\n   * mapped to the token type it lexes to. Delegates to the ExpressionLexer.\n   */\n  getKeywords(): Record<string, string> {\n    return this.expressionLexer.getKeywords();\n  }\n\n  next(): Token | undefined {\n    if (this.hasPeeked) {\n      this.hasPeeked = false;\n      return this.peekedToken;\n    }\n    // Materialized token array (ExpressionLexer path).\n    if (this.tokenIdx < this.tokens.length) {\n      return this.tokens[this.tokenIdx++];\n    }\n    return undefined;\n  }\n\n  peek(): Token | undefined {\n    if (this.hasPeeked) return this.peekedToken;\n    this.peekedToken = this.next();\n    this.hasPeeked = true;\n    return this.peekedToken;\n  }\n\n  [Symbol.iterator](): Iterator<Token> {\n    return this.tokens[Symbol.iterator]();\n  }\n\n  /**\n   * Register a plugin to extend the lexer with custom tokens.\n   * Delegates to the underlying ExpressionLexer.\n   *\n   * @see LexerVocabulary for the supported extension points.\n   */\n  registerVocabulary(plugin: LexerVocabulary): void {\n    this.expressionLexer.registerVocabulary(plugin);\n  }\n\n  /**\n   * Unregister a plugin, removing its custom tokens from the lexer.\n   * Delegates to the underlying ExpressionLexer.\n   */\n  unregisterVocabulary(plugin: LexerVocabulary): void {\n    this.expressionLexer.unregisterVocabulary(plugin);\n  }\n\n  /**\n   * Reset the lexer for expression-only text, skips the classifyLine()\n   * overhead in reset() for callers that already know the input is an\n   * evaluable expression (e.g., after isEmptyLine() confirmed non-skip).\n   */\n  resetExpression(input: string): void {\n    this.currentState = LexerState.Main;\n    this.hasPeeked = false;\n    this.peekedToken = undefined;\n    this.expressionLexer.reset(input);\n    this.tokens = this.expressionLexer.tokenizeAll();\n    this.tokenIdx = 0;\n  }\n\n  /**\n   * Scan a full document in one pass, classifying each line and\n   * tokenizing non-skipped lines. Delegates to ExpressionLexer.\n   *\n   * @returns ScanLineResult[], one per line, with classification + tokens.\n   */\n  scanDocument(text: string): ScanLineResult[] {\n    return this.expressionLexer.scanDocument(text);\n  }\n\n  getState(): LexerState {\n    return this.currentState;\n  }\n\n  setState(state: LexerState): void {\n    this.currentState = state;\n  }\n\n  getHighlightTokens(lineText: string): {type: string; value: string; offset: number; col: number; length: number; category: TokenCategory | undefined}[] {\n    const classification = this.expressionLexer.classifyLine(lineText);\n\n    // For blockquote lines, strip the \"> \" prefix and tokenize the expression content.\n    // This lets expressions inside blockquotes (e.g., \"> 1 + 2\") get syntax highlighted\n    // while pure structural lines (headings, code fences) remain unhighlighted.\n    if (classification.skip && lineText.startsWith(\"> \")) {\n      return this.collectHighlightTokens(lineText.slice(2));\n    }\n\n    if (classification.skip) {\n      return [];\n    }\n\n    return this.collectHighlightTokens(lineText);\n  }\n\n  /**\n   * The same tokens {@link getHighlightTokens} reduces, before reduction.\n   *\n   * Exists because normalization operates on tokens, not on the flattened\n   * shape, and a consumer that wants phrase-fused highlighting has to run the\n   * normalizer between the two. See `LanguageService.getSemanticTokens`.\n   *\n   * @param lineText - One line of source.\n   * @returns Every token on the line that is worth painting, unreduced.\n   */\n  getHighlightTokenObjects(lineText: string): Token[] {\n    const classification = this.expressionLexer.classifyLine(lineText);\n    if (classification.skip && lineText.startsWith(\"> \")) {\n      return this.collectTokenObjects(lineText.slice(2));\n    }\n    if (classification.skip) return [];\n    return this.collectTokenObjects(lineText);\n  }\n\n  private collectTokenObjects(lineText: string): Token[] {\n    this.resetExpression(lineText);\n    const result: Token[] = [];\n    for (const token of this) {\n      if (token.type === \"WS\" || token.type === \"NEWLINE\") continue;\n      if (token.type.startsWith(\"MD_\")) continue;\n      if (token.type === \"INLINE_SOLVE_START\" || token.type === \"BACKTICK_CLOSE\") continue;\n      result.push(token);\n    }\n    return result;\n  }\n\n  private collectHighlightTokens(lineText: string): {type: string; value: string; offset: number; col: number; length: number; category: TokenCategory | undefined}[] {\n    return this.collectTokenObjects(lineText).map(token => ({\n      type: token.type,\n      value: token.value,\n      offset: token.offset,\n      col: token.col,\n      // `text`, not `value`: this is a span into the source, and the two\n      // differ for a string literal, whose value is the payload while its\n      // text still carries the quote characters the reader typed.\n      length: token.text.length,\n      category: getTokenCategory(token.type),\n    }));\n  }\n}\n\n/**\n * A lexer for operations that do not depend on registered vocabulary.\n *\n * Line classification and inline-solve detection read characters looking for\n * headings, comment markers, fences and backtick spans, and never consult the\n * keyword, unit or operator tables. Every lexer therefore returns the same\n * answer, so the callers that have no engine to ask can use this one. Checked\n * by `__tests__/lexer/LineClassificationIsVocabularyIndependent.spec.ts`.\n *\n * Do not tokenize with this. An engine's own lexer carries the vocabulary its\n * packages registered; this one carries none.\n */\nexport const sharedLexer = new Lexer(\"en\", undefined);","/**\n * TokenClass, Plugin-extensible keyword registration for the Lexer.\n *\n * Providers call `registry.register(tokenClass)` to teach the lexer about\n * their keywords. The registry merges locale keywords, provider keywords,\n * phrase mappings, and unit names into an optimized TokenLookup structure\n * consumed by the Lexer.\n *\n * @example\n * registry.register({\n *   tokenType: 'CARET',\n *   keywords: {},\n *   phrases: { 'to the power of': true, 'power of': true },\n *   priority: 10,\n *   description: 'Exponentiation operators (x^y)',\n * });\n */\nexport interface TokenClass {\n  /** The token type string produced by the lexer (e.g., \"FUNC\", \"PI\", \"CARET\").\n   *  Must match a token type that a ParseletRegistry has a parselet for. */\n  tokenType: string;\n\n  /** Single-word keywords (case-insensitive). The lexer lowercases input\n   *  before lookup, so these should be lowercase. Example:\n   *  { sqrt: true, abs: true, sin: true, cos: true } for tokenType \"FUNC\" */\n  keywords: Record<string, boolean>;\n\n  /** Multi-word phrases (case-insensitive). Matched by the built-in PhraseMatcher\n   *  via the phrase trie. Example:\n   *  { \"to the power of\": true, \"power of\": true } for tokenType \"CARET\" */\n  phrases?: Record<string, boolean>;\n\n  /** Priority for conflict resolution. When two TokenClasses register\n   *  the same keyword, the higher-priority class wins. Locale keywords\n   *  have priority 0 (set via setLocale). Providers should use\n   *  priority >= 10 to override locale defaults. Default: 0 */\n  priority?: number;\n\n  /** Human-readable description for debugging and introspection */\n  description?: string;\n}\n\n// ── Phrase Trie ──────────────────────────────────────────────────────────────\n\n/** Trie node for multi-word phrase matching. */\nexport interface PhraseNode {\n  /** Complete phrase token type (null = intermediate node) */\n  type?: string;\n  children: Map<string, PhraseNode>;\n}\n\n// ── TokenLookup, Optimized lookup structure for the Lexer ──────────────────\n\n/**\n * The optimized lookup structure built by TokenClassRegistry.build().\n * Consumed by the Lexer for O(1) keyword → token type lookups and\n * O(word-count) phrase matching.\n */\nexport interface TokenLookup {\n  /** Lowercase keyword → token type. O(1) Map lookup. */\n  keywordToType: Map<string, string>;\n\n  /** Phrase trie for multi-word matching. Root node with children maps.\n   *  Null if no phrases registered. */\n  phraseTrie: PhraseNode | null;\n\n  /** Set of lowercase first-words of all registered phrases.\n   *  Used by the lexer to emit IDENT (not a phrase keyword) for words\n   *  that start multi-word phrases, deferring to the PhraseMatcher.\n   *\n   *  Example: \"to\" is in phraseStartWords because \"to the power of\" is a phrase.\n   *  When the lexer sees \"to\", it emits IDENT and lets the phrase matcher\n   *  combine \"to the power of\" into a single CARET token.\n   *\n   *  This prevents plugins from accidentally overriding phrase-start words.\n   */\n  phraseStartWords: Set<string>;\n\n  /** Case-sensitive unit names for UNIT fallback after keyword lookup fails. */\n  unitNames: ReadonlySet<string>;\n}\n\n// ── TokenClassRegistry ───────────────────────────────────────────────────────\n\n/**\n * Central registry for keyword→token-type mappings.\n *\n * Providers register TokenClasses; locales provide keyword maps;\n * units provide a name set. `build()` merges all sources into an\n * optimized TokenLookup consumed by the Lexer.\n *\n * Merge order (later overrides earlier):\n *   1. Locale keywords (priority 0)\n *   2. Provider keywords (sorted by priority ascending, higher priority wins)\n *\n * Unit names are stored separately (checked AFTER keyword lookup fails).\n * Phrases are stored in a trie for O(phrase-length) matching.\n */\nexport class TokenClassRegistry {\n  private classes: TokenClass[] = [];\n  private localeKeywordMap: Record<string, string> | null = null;\n  private localePhraseMap: Record<string, string> | null = null;\n  private unitNames: ReadonlySet<string> | null = null;\n\n  /**\n   * Register a provider's TokenClass. Must be called BEFORE build().\n   * Can be called multiple times to add more entries.\n   *\n   * Built-in token types CANNOT be overridden, throws a EngineError\n   * if the TokenClass attempts to register a keyword that conflicts\n   * with an already-registered token type.\n   */\n  register(tokenClass: TokenClass): void {\n    this.classes.push(tokenClass);\n  }\n\n  /**\n   * Unregister all TokenClasses for a given token type.\n   * Useful for plugin unload. Requires rebuild() to take effect.\n   */\n  unregister(tokenType: string): void {\n    this.classes = this.classes.filter(c => c.tokenType !== tokenType);\n  }\n\n  /**\n   * Set the locale's keyword→type map and optional phrase map.\n   * Called on locale change. Priority 0 (cannot override providers with higher priority).\n   */\n  setLocale(keywordMap: Record<string, string>, phraseMap?: Record<string, string>): void {\n    this.localeKeywordMap = keywordMap;\n    this.localePhraseMap = phraseMap ?? null;\n  }\n\n  /**\n   * Set the unit name set. Called when unit list changes.\n   * Units are stored separately (checked AFTER keyword lookup fails).\n   *\n   * Takes a ReadonlySet because the caller's set is derived from the\n   * conversion tables and must not be mutated; this class only ever reads it.\n   */\n  setUnits(unitNames: ReadonlySet<string>): void {\n    this.unitNames = unitNames;\n  }\n\n  /**\n   * Build the optimized TokenLookup from all registered sources.\n   *\n   * Merge order (later overrides earlier):\n   *   1. Locale keywords (priority 0)\n   *   2. Provider classes (sorted by priority ascending)\n   *\n   * Returns a frozen TokenLookup that the Lexer consumes.\n   * Call build() again after register()/setLocale()/setUnits() changes.\n   */\n  build(): TokenLookup {\n    const keywordToType = new Map<string, string>();\n\n    // Layer 1: Locale keywords (priority 0, lowest)\n    if (this.localeKeywordMap) {\n      for (const [keyword, tokenType] of Object.entries(this.localeKeywordMap)) {\n        keywordToType.set(keyword.toLowerCase(), tokenType);\n      }\n    }\n\n    // Layer 2: Provider keywords (sorted by priority ascending, higher wins)\n    const sorted = [...this.classes].sort((a, b) => (a.priority ?? 0) - (b.priority ?? 0));\n    for (const tc of sorted) {\n      for (const keyword of Object.keys(tc.keywords)) {\n        keywordToType.set(keyword.toLowerCase(), tc.tokenType);\n      }\n    }\n\n    // Build phrase trie from locale + provider phrases\n    const phraseTrie = this.buildPhraseTrie();\n\n    // Build phraseStartWords set (first word of every phrase)\n    const phraseStartWords = this.buildPhraseStartWords();\n\n    return {\n      keywordToType,\n      phraseTrie,\n      phraseStartWords,\n      unitNames: this.unitNames ?? new Set(),\n    };\n  }\n\n  // ── Private helpers ─────────────────────────────────────────────────────\n\n  private buildPhraseTrie(): PhraseNode | null {\n    const root: PhraseNode = { children: new Map() };\n\n    // Layer 1: Locale phrases\n    if (this.localePhraseMap) {\n      for (const [phrase, tokenType] of Object.entries(this.localePhraseMap)) {\n        this.insertPhrase(root, phrase.toLowerCase(), tokenType);\n      }\n    }\n\n    // Layer 2: Provider phrases\n    for (const tc of this.classes) {\n      if (!tc.phrases) continue;\n      for (const phrase of Object.keys(tc.phrases)) {\n        this.insertPhrase(root, phrase.toLowerCase(), tc.tokenType);\n      }\n    }\n\n    return root.children.size > 0 ? root : null;\n  }\n\n  private insertPhrase(root: PhraseNode, phrase: string, tokenType: string): void {\n    const words = phrase.split(' ');\n    let node = root;\n    for (const word of words) {\n      if (!node.children.has(word)) {\n        node.children.set(word, { children: new Map() });\n      }\n      node = node.children.get(word)!;\n    }\n    // Only set type if not already set, first-registered (locale) wins\n    if (!node.type) {\n      node.type = tokenType;\n    }\n  }\n\n  private buildPhraseStartWords(): Set<string> {\n    const startWords = new Set<string>();\n\n    // Locale phrase first words\n    if (this.localePhraseMap) {\n      for (const phrase of Object.keys(this.localePhraseMap)) {\n        const first = phrase.split(' ')[0].toLowerCase();\n        startWords.add(first);\n      }\n    }\n\n    // Provider phrase first words\n    for (const tc of this.classes) {\n      if (!tc.phrases) continue;\n      for (const phrase of Object.keys(tc.phrases)) {\n        const first = phrase.split(' ')[0].toLowerCase();\n        startWords.add(first);\n      }\n    }\n\n    return startWords;\n  }\n}","/**\n * tokenRegistration.ts, Bootstrap for building TokenLookup from locale, units, and phrases.\n *\n * This module centralizes the assembly of the TokenLookup consumed by ExpressionLexer.\n * It merges:\n *   1. Locale keywords (from ILocale.keywordMap)\n *   2. Built-in phrase patterns (to the power of, increase by, etc.)\n *   3. Known unit names (from units.ts)\n *\n * The resulting TokenLookup is frozen and passed to ExpressionLexer.configuredLookup\n * at construction time, enabling data-driven keyword/unit/phrase resolution.\n */\nimport { TokenClassRegistry } from '@solve-js/lexer/TokenClassRegistry';\nimport type { TokenLookup } from '@solve-js/lexer/TokenClassRegistry';\nimport { getLocale, type ILocale } from '@solve-js/constants/locales';\nimport { knownUnits } from '@solve-js/lexer/units';\n\n// ── Built-in phrase map ───────────────────────────────────────────────────\n// These are the multi-word expressions handled as compound tokens.\n// Matched by the PhraseMatcher (trie-based) in ExpressionLexer.tryMatchPhrase().\nconst BUILTIN_PHRASES: Record<string, string> = {\n  'to the power of': 'CARET',\n  'power of': 'CARET',\n  'increase by': 'INCREASE_BY',\n  'decrease by': 'DECREASE_BY',\n  'times by': 'TIMES_BY',\n  'multiply by': 'MULTIPLY_BY',\n  'multiplied by': 'MULTIPLY_BY',\n  'divide by': 'DIVIDE_BY',\n};\n\n/**\n * Build a TokenLookup from locale keywords, known units, and built-in phrases.\n *\n * Merge order (later overrides earlier):\n *   1. Locale keywords (priority 0, lowest, can be overridden by providers)\n *   2. Built-in phrases (via locale phraseMap)\n *   3. Known units (checked AFTER keyword lookup fails, via unitNames set)\n *\n * The resulting TokenLookup replaces the internal keyword map, unit set,\n * phrase trie, and phraseStartWords in ExpressionLexer when set via\n * ExpressionLexer.configuredLookup.\n *\n * @param localeCode - The locale code (e.g., \"en\", \"de\"). Defaults to \"en\".\n * @returns A frozen TokenLookup ready for consumption by the lexer.\n */\nexport function buildTokenLookup(localeCode = 'en'): TokenLookup {\n  const locale: ILocale = getLocale(localeCode);\n  const registry = new TokenClassRegistry();\n\n  // Layer 1: Locale keywords (priority 0, lowest)\n  registry.setLocale(locale.keywordMap, BUILTIN_PHRASES);\n\n  // Layer 2: Known units (checked after keyword lookup)\n  registry.setUnits(knownUnits);\n\n  // Build and return the lookup\n  return registry.build();\n}\n"]}
+{"version":3,"sources":["../src/lexer/LexerState.ts","../src/lexer/Lexer.ts","../src/lexer/TokenClassRegistry.ts","../src/lexer/tokenRegistration.ts"],"names":["LexerState","Lexer","localeCode","tokenLookup","ExpressionLexer","input","state","newState","lineText","plugin","text","classification","result","token","getTokenCategory","sharedLexer","TokenClassRegistry","tokenClass","tokenType","c","keywordMap","phraseMap","unitNames","keywordToType","keyword","sorted","a","b","tc","phraseTrie","phraseStartWords","root","phrase","words","node","word","startWords","first","BUILTIN_PHRASES","buildTokenLookup","locale","getLocale","registry","knownUnits"],"mappings":"uJAMO,IAAKA,CAAAA,CAAAA,CAAAA,CAAAA,GACXA,EAAA,IAAA,CAAO,MAAA,CACPA,EAAA,MAAA,CAAS,QAAA,CACTA,EAAA,MAAA,CAAS,QAAA,CAHEA,OAAA,EAAA,ECaL,IAAMC,EAAN,KAAY,CAkBjB,YAAYC,CAAAA,CAAa,IAAA,CAAMC,EAA2B,CAf1D,IAAA,CAAQ,aAA2B,MAAA,CAEnC,IAAA,CAAQ,UAAY,KAAA,CAIpB,IAAA,CAAQ,OAAkB,EAAC,CAC3B,KAAQ,QAAA,CAAmB,CAAA,CAYzB,KAAK,eAAA,CAAkB,IAAIC,EAAgBF,CAAAA,CAAYC,CAAW,EACpE,CAEA,KAAA,CAAME,EAAeC,CAAAA,CAA0B,CAC7C,IAAMC,CAAAA,CAAWD,CAAAA,EAAS,OAQ1B,GAPA,IAAA,CAAK,aAAeC,CAAAA,CACpB,IAAA,CAAK,UAAY,KAAA,CACjB,IAAA,CAAK,YAAc,MAAA,CAKfA,CAAAA,GAAa,OAAiB,CAEhC,GADuB,KAAK,eAAA,CAAgB,YAAA,CAAaF,CAAK,CAAA,CAC3C,IAAA,CAAM,CACvB,IAAA,CAAK,MAAA,CAAS,EAAC,CACf,IAAA,CAAK,SAAW,CAAA,CAChB,MACF,CAEA,IAAA,CAAK,eAAA,CAAgB,MAAMA,CAAK,CAAA,CAChC,KAAK,MAAA,CAAS,IAAA,CAAK,gBAAgB,WAAA,EAAY,CAC/C,KAAK,QAAA,CAAW,EAClB,MAEE,IAAA,CAAK,eAAA,CAAgB,MAAMA,CAAK,CAAA,CAChC,KAAK,MAAA,CAAS,IAAA,CAAK,gBAAgB,WAAA,EAAY,CAC/C,KAAK,QAAA,CAAW,EAEpB,CAMA,YAAA,CAAaG,CAAAA,CAAsC,CACjD,OAAO,IAAA,CAAK,gBAAgB,YAAA,CAAaA,CAAQ,CACnD,CAMA,gBAAA,CAAiBA,EAAkB,CACjC,OAAO,KAAK,eAAA,CAAgB,gBAAA,CAAiBA,CAAQ,CACvD,CAMA,aAAsC,CACpC,OAAO,KAAK,eAAA,CAAgB,WAAA,EAC9B,CAEA,IAAA,EAA0B,CACxB,GAAI,IAAA,CAAK,UACP,OAAA,IAAA,CAAK,SAAA,CAAY,MACV,IAAA,CAAK,WAAA,CAGd,GAAI,IAAA,CAAK,QAAA,CAAW,KAAK,MAAA,CAAO,MAAA,CAC9B,OAAO,IAAA,CAAK,MAAA,CAAO,KAAK,QAAA,EAAU,CAGtC,CAEA,IAAA,EAA0B,CACxB,OAAI,IAAA,CAAK,SAAA,CAAkB,KAAK,WAAA,EAChC,IAAA,CAAK,YAAc,IAAA,CAAK,IAAA,GACxB,IAAA,CAAK,SAAA,CAAY,KACV,IAAA,CAAK,WAAA,CACd,CAEA,CAAC,MAAA,CAAO,QAAQ,CAAA,EAAqB,CACnC,OAAO,IAAA,CAAK,MAAA,CAAO,MAAA,CAAO,QAAQ,CAAA,EACpC,CAQA,kBAAA,CAAmBC,CAAAA,CAA+B,CAChD,IAAA,CAAK,eAAA,CAAgB,mBAAmBA,CAAM,EAChD,CAMA,oBAAA,CAAqBA,CAAAA,CAA+B,CAClD,IAAA,CAAK,eAAA,CAAgB,qBAAqBA,CAAM,EAClD,CAOA,eAAA,CAAgBJ,CAAAA,CAAqB,CACnC,IAAA,CAAK,YAAA,CAAe,OACpB,IAAA,CAAK,SAAA,CAAY,MACjB,IAAA,CAAK,WAAA,CAAc,OACnB,IAAA,CAAK,eAAA,CAAgB,MAAMA,CAAK,CAAA,CAChC,KAAK,MAAA,CAAS,IAAA,CAAK,gBAAgB,WAAA,EAAY,CAC/C,KAAK,QAAA,CAAW,EAClB,CAQA,YAAA,CAAaK,CAAAA,CAAgC,CAC3C,OAAO,IAAA,CAAK,gBAAgB,YAAA,CAAaA,CAAI,CAC/C,CAEA,QAAA,EAAuB,CACrB,OAAO,IAAA,CAAK,YACd,CAEA,QAAA,CAASJ,EAAyB,CAChC,IAAA,CAAK,aAAeA,EACtB,CAEA,mBAAmBE,CAAAA,CAAqI,CACtJ,IAAMG,CAAAA,CAAiB,IAAA,CAAK,gBAAgB,YAAA,CAAaH,CAAQ,EAKjE,OAAIG,CAAAA,CAAe,MAAQH,CAAAA,CAAS,UAAA,CAAW,IAAI,CAAA,CAC1C,IAAA,CAAK,uBAAuBA,CAAAA,CAAS,KAAA,CAAM,CAAC,CAAC,CAAA,CAGlDG,EAAe,IAAA,CACV,GAGF,IAAA,CAAK,sBAAA,CAAuBH,CAAQ,CAC7C,CAYA,yBAAyBA,CAAAA,CAA2B,CAClD,IAAMG,CAAAA,CAAiB,IAAA,CAAK,gBAAgB,YAAA,CAAaH,CAAQ,EACjE,OAAIG,CAAAA,CAAe,MAAQH,CAAAA,CAAS,UAAA,CAAW,IAAI,CAAA,CAC1C,IAAA,CAAK,oBAAoBA,CAAAA,CAAS,KAAA,CAAM,CAAC,CAAC,CAAA,CAE/CG,EAAe,IAAA,CAAa,GACzB,IAAA,CAAK,mBAAA,CAAoBH,CAAQ,CAC1C,CAEQ,oBAAoBA,CAAAA,CAA2B,CACrD,KAAK,eAAA,CAAgBA,CAAQ,EAC7B,IAAMI,CAAAA,CAAkB,EAAC,CACzB,IAAA,IAAWC,KAAS,IAAA,CACdA,CAAAA,CAAM,OAAS,IAAA,EAAQA,CAAAA,CAAM,OAAS,SAAA,EACtCA,CAAAA,CAAM,KAAK,UAAA,CAAW,KAAK,GAC3BA,CAAAA,CAAM,IAAA,GAAS,sBAAwBA,CAAAA,CAAM,IAAA,GAAS,kBAC1DD,CAAAA,CAAO,IAAA,CAAKC,CAAK,CAAA,CAEnB,OAAOD,CACT,CAEQ,sBAAA,CAAuBJ,EAAqI,CAClK,OAAO,KAAK,mBAAA,CAAoBA,CAAQ,EAAE,GAAA,CAAIK,CAAAA,GAAU,CACtD,IAAA,CAAMA,CAAAA,CAAM,KACZ,KAAA,CAAOA,CAAAA,CAAM,KAAA,CACb,MAAA,CAAQA,CAAAA,CAAM,MAAA,CACd,IAAKA,CAAAA,CAAM,GAAA,CAIX,OAAQA,CAAAA,CAAM,IAAA,CAAK,OACnB,QAAA,CAAUC,CAAAA,CAAiBD,EAAM,IAAI,CACvC,EAAE,CACJ,CACF,EAcaE,CAAAA,CAAc,IAAId,EAAM,IAAA,CAAM,MAAS,EClJ7C,IAAMe,CAAAA,CAAN,KAAyB,CAAzB,WAAA,EAAA,CACL,KAAQ,OAAA,CAAwB,GAChC,IAAA,CAAQ,gBAAA,CAAkD,KAC1D,IAAA,CAAQ,eAAA,CAAiD,KACzD,IAAA,CAAQ,SAAA,CAAwC,MAUhD,QAAA,CAASC,CAAAA,CAA8B,CACrC,IAAA,CAAK,OAAA,CAAQ,KAAKA,CAAU,EAC9B,CAMA,UAAA,CAAWC,CAAAA,CAAyB,CAClC,IAAA,CAAK,OAAA,CAAU,KAAK,OAAA,CAAQ,MAAA,CAAOC,GAAKA,CAAAA,CAAE,SAAA,GAAcD,CAAS,EACnE,CAMA,UAAUE,CAAAA,CAAoCC,CAAAA,CAA0C,CACtF,IAAA,CAAK,gBAAA,CAAmBD,EACxB,IAAA,CAAK,eAAA,CAAkBC,GAAa,KACtC,CASA,SAASC,CAAAA,CAAsC,CAC7C,KAAK,SAAA,CAAYA,EACnB,CAYA,KAAA,EAAqB,CACnB,IAAMC,CAAAA,CAAgB,IAAI,IAG1B,GAAI,IAAA,CAAK,iBACP,IAAA,GAAW,CAACC,EAASN,CAAS,CAAA,GAAK,OAAO,OAAA,CAAQ,IAAA,CAAK,gBAAgB,CAAA,CACrEK,CAAAA,CAAc,IAAIC,CAAAA,CAAQ,WAAA,GAAeN,CAAS,CAAA,CAKtD,IAAMO,CAAAA,CAAS,CAAC,GAAG,IAAA,CAAK,OAAO,EAAE,IAAA,CAAK,CAACC,EAAGC,CAAAA,GAAAA,CAAOD,CAAAA,CAAE,UAAY,CAAA,GAAMC,CAAAA,CAAE,UAAY,CAAA,CAAE,CAAA,CACrF,QAAWC,CAAAA,IAAMH,CAAAA,CACf,QAAWD,CAAAA,IAAW,MAAA,CAAO,KAAKI,CAAAA,CAAG,QAAQ,EAC3CL,CAAAA,CAAc,GAAA,CAAIC,EAAQ,WAAA,EAAY,CAAGI,EAAG,SAAS,CAAA,CAKzD,IAAMC,CAAAA,CAAa,IAAA,CAAK,iBAAgB,CAGlCC,CAAAA,CAAmB,KAAK,qBAAA,EAAsB,CAEpD,OAAO,CACL,aAAA,CAAAP,EACA,UAAA,CAAAM,CAAAA,CACA,iBAAAC,CAAAA,CACA,SAAA,CAAW,KAAK,SAAA,EAAa,IAAI,GACnC,CACF,CAIQ,iBAAqC,CAC3C,IAAMC,EAAmB,CAAE,QAAA,CAAU,IAAI,GAAM,CAAA,CAG/C,GAAI,IAAA,CAAK,eAAA,CACP,OAAW,CAACC,CAAAA,CAAQd,CAAS,CAAA,GAAK,MAAA,CAAO,QAAQ,IAAA,CAAK,eAAe,CAAA,CACnE,IAAA,CAAK,YAAA,CAAaa,CAAAA,CAAMC,EAAO,WAAA,EAAY,CAAGd,CAAS,CAAA,CAK3D,IAAA,IAAWU,KAAM,IAAA,CAAK,OAAA,CACpB,GAAKA,CAAAA,CAAG,OAAA,CACR,QAAWI,CAAAA,IAAU,MAAA,CAAO,KAAKJ,CAAAA,CAAG,OAAO,EACzC,IAAA,CAAK,YAAA,CAAaG,EAAMC,CAAAA,CAAO,WAAA,GAAeJ,CAAAA,CAAG,SAAS,EAI9D,OAAOG,CAAAA,CAAK,SAAS,IAAA,CAAO,CAAA,CAAIA,EAAO,IACzC,CAEQ,aAAaA,CAAAA,CAAkBC,CAAAA,CAAgBd,EAAyB,CAC9E,IAAMe,EAAQD,CAAAA,CAAO,KAAA,CAAM,GAAG,CAAA,CAC1BE,CAAAA,CAAOH,EACX,IAAA,IAAWI,CAAAA,IAAQF,EACZC,CAAAA,CAAK,QAAA,CAAS,IAAIC,CAAI,CAAA,EACzBD,EAAK,QAAA,CAAS,GAAA,CAAIC,EAAM,CAAE,QAAA,CAAU,IAAI,GAAM,CAAC,EAEjDD,CAAAA,CAAOA,CAAAA,CAAK,SAAS,GAAA,CAAIC,CAAI,EAG1BD,CAAAA,CAAK,IAAA,GACRA,EAAK,IAAA,CAAOhB,CAAAA,EAEhB,CAEQ,qBAAA,EAAqC,CAC3C,IAAMkB,CAAAA,CAAa,IAAI,IAGvB,GAAI,IAAA,CAAK,gBACP,IAAA,IAAWJ,CAAAA,IAAU,OAAO,IAAA,CAAK,IAAA,CAAK,eAAe,CAAA,CAAG,CACtD,IAAMK,CAAAA,CAAQL,CAAAA,CAAO,MAAM,GAAG,CAAA,CAAE,CAAC,CAAA,CAAE,WAAA,GACnCI,CAAAA,CAAW,GAAA,CAAIC,CAAK,EACtB,CAIF,QAAWT,CAAAA,IAAM,IAAA,CAAK,QACpB,GAAKA,CAAAA,CAAG,QACR,IAAA,IAAWI,CAAAA,IAAU,OAAO,IAAA,CAAKJ,CAAAA,CAAG,OAAO,CAAA,CAAG,CAC5C,IAAMS,CAAAA,CAAQL,CAAAA,CAAO,MAAM,GAAG,CAAA,CAAE,CAAC,CAAA,CAAE,WAAA,GACnCI,CAAAA,CAAW,GAAA,CAAIC,CAAK,EACtB,CAGF,OAAOD,CACT,CACF,EClOA,IAAME,CAAAA,CAA0C,CAC9C,iBAAA,CAAmB,OAAA,CACnB,WAAY,OAAA,CACZ,aAAA,CAAe,cACf,aAAA,CAAe,aAAA,CACf,WAAY,UAAA,CACZ,aAAA,CAAe,cACf,eAAA,CAAiB,aAAA,CACjB,YAAa,WACf,CAAA,CAiBO,SAASC,CAAAA,CAAiBrC,GAAAA,CAAa,KAAmB,CAC/D,IAAMsC,EAAkBC,CAAAA,CAAUvC,GAAU,EACtCwC,CAAAA,CAAW,IAAI1B,EAGrB,OAAA0B,CAAAA,CAAS,UAAUF,CAAAA,CAAO,UAAA,CAAYF,CAAe,CAAA,CAGrDI,CAAAA,CAAS,SAASC,GAAU,CAAA,CAGrBD,CAAAA,CAAS,KAAA,EAClB","file":"chunk-2DPKJ2SM.js","sourcesContent":["/**\n * Lexer state machine modes.\n * - Main: document-level scanning with markdown classification\n * - Inline: expression embedded in markdown inline solve (`s\\`...\\``)\n * - String: inside a double-quoted string literal\n */\nexport enum LexerState {\n\tMain = \"main\",\n\tInline = \"inline\",\n\tString = \"string\",\n}\n","import { ExpressionLexer, LineClassification, LexerVocabulary, type ScanLineResult } from \"./ExpressionLexer\";\nimport { Token } from \"@solve-js/lexer/Token\";\nimport { LexerState } from \"@solve-js/lexer/LexerState\";\nimport { getTokenCategory } from \"@solve-js/language/TokenCategoryMap\";\nimport type { TokenCategory } from \"@solve-js/language/TokenCategory\";\nimport type { TokenLookup } from \"@solve-js/lexer/TokenClassRegistry\";\n\n/**\n * Public tokenizer wrapper around {@link ExpressionLexer}.\n *\n * `ExpressionLexer` does the actual character-by-character scanning;\n * `Lexer` adds a materialized-token-array streaming interface\n * (`next()`/`peek()`) plus line-classification state (`reset()`) so\n * callers can iterate a line's tokens without re-scanning on each peek.\n *\n * Each `ExpressionEngine` instance owns its own `Lexer`, and packages\n * extend it via {@link registerVocabulary} (keywords, operators, units)\n * see `IEnginePackage.lexerVocabulary`.\n */\nexport class Lexer {\n  /** Expression-mode lexer (Phase A: V8-optimized, replaces moo) */\n  private expressionLexer: ExpressionLexer;\n  private currentState: LexerState = LexerState.Main;\n  private peekedToken: Token | undefined;\n  private hasPeeked = false;\n\n  // Materialized token array from the last reset() call, used for\n  // next()/peek() streaming access.\n  private tokens: Token[] = [];\n  private tokenIdx: number = 0;\n\n  /**\n   * @param localeCode - Locale code (e.g., \"en\", \"de\"). Defaults to \"en\".\n   * @param tokenLookup - Optional TokenLookup from TokenClassRegistry.\n   *   When provided, configures ExpressionLexer to use registry-built\n   *   keyword/unit/phrase lookups instead of internal instance maps.\n   */\n  constructor(localeCode = \"en\", tokenLookup?: TokenLookup) {\n    // Pass the lookup directly to ExpressionLexer's constructor, it's an\n    // instance field now, not a static. Each Lexer instance gets its own\n    // isolated lookup, preventing cross-instance corruption.\n    this.expressionLexer = new ExpressionLexer(localeCode, tokenLookup);\n  }\n\n  reset(input: string, state?: LexerState): void {\n    const newState = state ?? LexerState.Main;\n    this.currentState = newState;\n    this.hasPeeked = false;\n    this.peekedToken = undefined;\n\n    // Phase B: Main state classifies the line with the markdown scanner.\n    // Skip lines (headings, fences, HRs, etc.) produce empty token arrays.\n    // Expression lines and lines with inline solves are tokenized normally.\n    if (newState === LexerState.Main) {\n      const classification = this.expressionLexer.classifyLine(input);\n      if (classification.skip) {\n        this.tokens = [];\n        this.tokenIdx = 0;\n        return;\n      }\n      // Expression line or markdown line with inline solves, tokenize.\n      this.expressionLexer.reset(input);\n      this.tokens = this.expressionLexer.tokenizeAll();\n      this.tokenIdx = 0;\n    } else {\n      // Non-main states (Inline, String), expression tokenization.\n      this.expressionLexer.reset(input);\n      this.tokens = this.expressionLexer.tokenizeAll();\n      this.tokenIdx = 0;\n    }\n  }\n\n  /**\n   * Classify a single line of markdown text (Phase B).\n   * Delegates to the ExpressionLexer's character-by-character scanner.\n   */\n  classifyLine(lineText: string): LineClassification {\n    return this.expressionLexer.classifyLine(lineText);\n  }\n\n  /**\n   * Find all inline solve markers in a line (Phase B).\n   * Delegates to the ExpressionLexer's character-by-character scanner.\n   */\n  findInlineSolves(lineText: string) {\n    return this.expressionLexer.findInlineSolves(lineText);\n  }\n\n  /**\n   * Every keyword this lexer currently recognizes (locale + plugin-contributed),\n   * mapped to the token type it lexes to. Delegates to the ExpressionLexer.\n   */\n  getKeywords(): Record<string, string> {\n    return this.expressionLexer.getKeywords();\n  }\n\n  next(): Token | undefined {\n    if (this.hasPeeked) {\n      this.hasPeeked = false;\n      return this.peekedToken;\n    }\n    // Materialized token array (ExpressionLexer path).\n    if (this.tokenIdx < this.tokens.length) {\n      return this.tokens[this.tokenIdx++];\n    }\n    return undefined;\n  }\n\n  peek(): Token | undefined {\n    if (this.hasPeeked) return this.peekedToken;\n    this.peekedToken = this.next();\n    this.hasPeeked = true;\n    return this.peekedToken;\n  }\n\n  [Symbol.iterator](): Iterator<Token> {\n    return this.tokens[Symbol.iterator]();\n  }\n\n  /**\n   * Register a plugin to extend the lexer with custom tokens.\n   * Delegates to the underlying ExpressionLexer.\n   *\n   * @see LexerVocabulary for the supported extension points.\n   */\n  registerVocabulary(plugin: LexerVocabulary): void {\n    this.expressionLexer.registerVocabulary(plugin);\n  }\n\n  /**\n   * Unregister a plugin, removing its custom tokens from the lexer.\n   * Delegates to the underlying ExpressionLexer.\n   */\n  unregisterVocabulary(plugin: LexerVocabulary): void {\n    this.expressionLexer.unregisterVocabulary(plugin);\n  }\n\n  /**\n   * Reset the lexer for expression-only text, skips the classifyLine()\n   * overhead in reset() for callers that already know the input is an\n   * evaluable expression (e.g., after isEmptyLine() confirmed non-skip).\n   */\n  resetExpression(input: string): void {\n    this.currentState = LexerState.Main;\n    this.hasPeeked = false;\n    this.peekedToken = undefined;\n    this.expressionLexer.reset(input);\n    this.tokens = this.expressionLexer.tokenizeAll();\n    this.tokenIdx = 0;\n  }\n\n  /**\n   * Scan a full document in one pass, classifying each line and\n   * tokenizing non-skipped lines. Delegates to ExpressionLexer.\n   *\n   * @returns ScanLineResult[], one per line, with classification + tokens.\n   */\n  scanDocument(text: string): ScanLineResult[] {\n    return this.expressionLexer.scanDocument(text);\n  }\n\n  getState(): LexerState {\n    return this.currentState;\n  }\n\n  setState(state: LexerState): void {\n    this.currentState = state;\n  }\n\n  getHighlightTokens(lineText: string): {type: string; value: string; offset: number; col: number; length: number; category: TokenCategory | undefined}[] {\n    const classification = this.expressionLexer.classifyLine(lineText);\n\n    // For blockquote lines, strip the \"> \" prefix and tokenize the expression content.\n    // This lets expressions inside blockquotes (e.g., \"> 1 + 2\") get syntax highlighted\n    // while pure structural lines (headings, code fences) remain unhighlighted.\n    if (classification.skip && lineText.startsWith(\"> \")) {\n      return this.collectHighlightTokens(lineText.slice(2));\n    }\n\n    if (classification.skip) {\n      return [];\n    }\n\n    return this.collectHighlightTokens(lineText);\n  }\n\n  /**\n   * The same tokens {@link getHighlightTokens} reduces, before reduction.\n   *\n   * Exists because normalization operates on tokens, not on the flattened\n   * shape, and a consumer that wants phrase-fused highlighting has to run the\n   * normalizer between the two. See `LanguageService.getSemanticTokens`.\n   *\n   * @param lineText - One line of source.\n   * @returns Every token on the line that is worth painting, unreduced.\n   */\n  getHighlightTokenObjects(lineText: string): Token[] {\n    const classification = this.expressionLexer.classifyLine(lineText);\n    if (classification.skip && lineText.startsWith(\"> \")) {\n      return this.collectTokenObjects(lineText.slice(2));\n    }\n    if (classification.skip) return [];\n    return this.collectTokenObjects(lineText);\n  }\n\n  private collectTokenObjects(lineText: string): Token[] {\n    this.resetExpression(lineText);\n    const result: Token[] = [];\n    for (const token of this) {\n      if (token.type === \"WS\" || token.type === \"NEWLINE\") continue;\n      if (token.type.startsWith(\"MD_\")) continue;\n      if (token.type === \"INLINE_SOLVE_START\" || token.type === \"BACKTICK_CLOSE\") continue;\n      result.push(token);\n    }\n    return result;\n  }\n\n  private collectHighlightTokens(lineText: string): {type: string; value: string; offset: number; col: number; length: number; category: TokenCategory | undefined}[] {\n    return this.collectTokenObjects(lineText).map(token => ({\n      type: token.type,\n      value: token.value,\n      offset: token.offset,\n      col: token.col,\n      // `text`, not `value`: this is a span into the source, and the two\n      // differ for a string literal, whose value is the payload while its\n      // text still carries the quote characters the reader typed.\n      length: token.text.length,\n      category: getTokenCategory(token.type),\n    }));\n  }\n}\n\n/**\n * A lexer for operations that do not depend on registered vocabulary.\n *\n * Line classification and inline-solve detection read characters looking for\n * headings, comment markers, fences and backtick spans, and never consult the\n * keyword, unit or operator tables. Every lexer therefore returns the same\n * answer, so the callers that have no engine to ask can use this one. Checked\n * by `__tests__/lexer/LineClassificationIsVocabularyIndependent.spec.ts`.\n *\n * Do not tokenize with this. An engine's own lexer carries the vocabulary its\n * packages registered; this one carries none.\n */\nexport const sharedLexer = new Lexer(\"en\", undefined);","/**\n * TokenClass, Plugin-extensible keyword registration for the Lexer.\n *\n * Providers call `registry.register(tokenClass)` to teach the lexer about\n * their keywords. The registry merges locale keywords, provider keywords,\n * phrase mappings, and unit names into an optimized TokenLookup structure\n * consumed by the Lexer.\n *\n * @example\n * registry.register({\n *   tokenType: 'CARET',\n *   keywords: {},\n *   phrases: { 'to the power of': true, 'power of': true },\n *   priority: 10,\n *   description: 'Exponentiation operators (x^y)',\n * });\n */\nexport interface TokenClass {\n  /** The token type string produced by the lexer (e.g., \"FUNC\", \"PI\", \"CARET\").\n   *  Must match a token type that a ParseletRegistry has a parselet for. */\n  tokenType: string;\n\n  /** Single-word keywords (case-insensitive). The lexer lowercases input\n   *  before lookup, so these should be lowercase. Example:\n   *  { sqrt: true, abs: true, sin: true, cos: true } for tokenType \"FUNC\" */\n  keywords: Record<string, boolean>;\n\n  /** Multi-word phrases (case-insensitive). Matched by the built-in PhraseMatcher\n   *  via the phrase trie. Example:\n   *  { \"to the power of\": true, \"power of\": true } for tokenType \"CARET\" */\n  phrases?: Record<string, boolean>;\n\n  /** Priority for conflict resolution. When two TokenClasses register\n   *  the same keyword, the higher-priority class wins. Locale keywords\n   *  have priority 0 (set via setLocale). Providers should use\n   *  priority >= 10 to override locale defaults. Default: 0 */\n  priority?: number;\n\n  /** Human-readable description for debugging and introspection */\n  description?: string;\n}\n\n// ── Phrase Trie ──────────────────────────────────────────────────────────────\n\n/** Trie node for multi-word phrase matching. */\nexport interface PhraseNode {\n  /** Complete phrase token type (null = intermediate node) */\n  type?: string;\n  children: Map<string, PhraseNode>;\n}\n\n// ── TokenLookup, Optimized lookup structure for the Lexer ──────────────────\n\n/**\n * The optimized lookup structure built by TokenClassRegistry.build().\n * Consumed by the Lexer for O(1) keyword → token type lookups and\n * O(word-count) phrase matching.\n */\nexport interface TokenLookup {\n  /** Lowercase keyword → token type. O(1) Map lookup. */\n  keywordToType: Map<string, string>;\n\n  /** Phrase trie for multi-word matching. Root node with children maps.\n   *  Null if no phrases registered. */\n  phraseTrie: PhraseNode | null;\n\n  /** Set of lowercase first-words of all registered phrases.\n   *  Used by the lexer to emit IDENT (not a phrase keyword) for words\n   *  that start multi-word phrases, deferring to the PhraseMatcher.\n   *\n   *  Example: \"to\" is in phraseStartWords because \"to the power of\" is a phrase.\n   *  When the lexer sees \"to\", it emits IDENT and lets the phrase matcher\n   *  combine \"to the power of\" into a single CARET token.\n   *\n   *  This prevents plugins from accidentally overriding phrase-start words.\n   */\n  phraseStartWords: Set<string>;\n\n  /** Case-sensitive unit names for UNIT fallback after keyword lookup fails. */\n  unitNames: ReadonlySet<string>;\n}\n\n// ── TokenClassRegistry ───────────────────────────────────────────────────────\n\n/**\n * Central registry for keyword→token-type mappings.\n *\n * Providers register TokenClasses; locales provide keyword maps;\n * units provide a name set. `build()` merges all sources into an\n * optimized TokenLookup consumed by the Lexer.\n *\n * Merge order (later overrides earlier):\n *   1. Locale keywords (priority 0)\n *   2. Provider keywords (sorted by priority ascending, higher priority wins)\n *\n * Unit names are stored separately (checked AFTER keyword lookup fails).\n * Phrases are stored in a trie for O(phrase-length) matching.\n */\nexport class TokenClassRegistry {\n  private classes: TokenClass[] = [];\n  private localeKeywordMap: Record<string, string> | null = null;\n  private localePhraseMap: Record<string, string> | null = null;\n  private unitNames: ReadonlySet<string> | null = null;\n\n  /**\n   * Register a provider's TokenClass. Must be called BEFORE build().\n   * Can be called multiple times to add more entries.\n   *\n   * Built-in token types CANNOT be overridden, throws a EngineError\n   * if the TokenClass attempts to register a keyword that conflicts\n   * with an already-registered token type.\n   */\n  register(tokenClass: TokenClass): void {\n    this.classes.push(tokenClass);\n  }\n\n  /**\n   * Unregister all TokenClasses for a given token type.\n   * Useful for plugin unload. Requires rebuild() to take effect.\n   */\n  unregister(tokenType: string): void {\n    this.classes = this.classes.filter(c => c.tokenType !== tokenType);\n  }\n\n  /**\n   * Set the locale's keyword→type map and optional phrase map.\n   * Called on locale change. Priority 0 (cannot override providers with higher priority).\n   */\n  setLocale(keywordMap: Record<string, string>, phraseMap?: Record<string, string>): void {\n    this.localeKeywordMap = keywordMap;\n    this.localePhraseMap = phraseMap ?? null;\n  }\n\n  /**\n   * Set the unit name set. Called when unit list changes.\n   * Units are stored separately (checked AFTER keyword lookup fails).\n   *\n   * Takes a ReadonlySet because the caller's set is derived from the\n   * conversion tables and must not be mutated; this class only ever reads it.\n   */\n  setUnits(unitNames: ReadonlySet<string>): void {\n    this.unitNames = unitNames;\n  }\n\n  /**\n   * Build the optimized TokenLookup from all registered sources.\n   *\n   * Merge order (later overrides earlier):\n   *   1. Locale keywords (priority 0)\n   *   2. Provider classes (sorted by priority ascending)\n   *\n   * Returns a frozen TokenLookup that the Lexer consumes.\n   * Call build() again after register()/setLocale()/setUnits() changes.\n   */\n  build(): TokenLookup {\n    const keywordToType = new Map<string, string>();\n\n    // Layer 1: Locale keywords (priority 0, lowest)\n    if (this.localeKeywordMap) {\n      for (const [keyword, tokenType] of Object.entries(this.localeKeywordMap)) {\n        keywordToType.set(keyword.toLowerCase(), tokenType);\n      }\n    }\n\n    // Layer 2: Provider keywords (sorted by priority ascending, higher wins)\n    const sorted = [...this.classes].sort((a, b) => (a.priority ?? 0) - (b.priority ?? 0));\n    for (const tc of sorted) {\n      for (const keyword of Object.keys(tc.keywords)) {\n        keywordToType.set(keyword.toLowerCase(), tc.tokenType);\n      }\n    }\n\n    // Build phrase trie from locale + provider phrases\n    const phraseTrie = this.buildPhraseTrie();\n\n    // Build phraseStartWords set (first word of every phrase)\n    const phraseStartWords = this.buildPhraseStartWords();\n\n    return {\n      keywordToType,\n      phraseTrie,\n      phraseStartWords,\n      unitNames: this.unitNames ?? new Set(),\n    };\n  }\n\n  // ── Private helpers ─────────────────────────────────────────────────────\n\n  private buildPhraseTrie(): PhraseNode | null {\n    const root: PhraseNode = { children: new Map() };\n\n    // Layer 1: Locale phrases\n    if (this.localePhraseMap) {\n      for (const [phrase, tokenType] of Object.entries(this.localePhraseMap)) {\n        this.insertPhrase(root, phrase.toLowerCase(), tokenType);\n      }\n    }\n\n    // Layer 2: Provider phrases\n    for (const tc of this.classes) {\n      if (!tc.phrases) continue;\n      for (const phrase of Object.keys(tc.phrases)) {\n        this.insertPhrase(root, phrase.toLowerCase(), tc.tokenType);\n      }\n    }\n\n    return root.children.size > 0 ? root : null;\n  }\n\n  private insertPhrase(root: PhraseNode, phrase: string, tokenType: string): void {\n    const words = phrase.split(' ');\n    let node = root;\n    for (const word of words) {\n      if (!node.children.has(word)) {\n        node.children.set(word, { children: new Map() });\n      }\n      node = node.children.get(word)!;\n    }\n    // Only set type if not already set, first-registered (locale) wins\n    if (!node.type) {\n      node.type = tokenType;\n    }\n  }\n\n  private buildPhraseStartWords(): Set<string> {\n    const startWords = new Set<string>();\n\n    // Locale phrase first words\n    if (this.localePhraseMap) {\n      for (const phrase of Object.keys(this.localePhraseMap)) {\n        const first = phrase.split(' ')[0].toLowerCase();\n        startWords.add(first);\n      }\n    }\n\n    // Provider phrase first words\n    for (const tc of this.classes) {\n      if (!tc.phrases) continue;\n      for (const phrase of Object.keys(tc.phrases)) {\n        const first = phrase.split(' ')[0].toLowerCase();\n        startWords.add(first);\n      }\n    }\n\n    return startWords;\n  }\n}","/**\n * tokenRegistration.ts, Bootstrap for building TokenLookup from locale, units, and phrases.\n *\n * This module centralizes the assembly of the TokenLookup consumed by ExpressionLexer.\n * It merges:\n *   1. Locale keywords (from ILocale.keywordMap)\n *   2. Built-in phrase patterns (to the power of, increase by, etc.)\n *   3. Known unit names (from units.ts)\n *\n * The resulting TokenLookup is frozen and passed to ExpressionLexer.configuredLookup\n * at construction time, enabling data-driven keyword/unit/phrase resolution.\n */\nimport { TokenClassRegistry } from '@solve-js/lexer/TokenClassRegistry';\nimport type { TokenLookup } from '@solve-js/lexer/TokenClassRegistry';\nimport { getLocale, type ILocale } from '@solve-js/constants/locales';\nimport { knownUnits } from '@solve-js/lexer/units';\n\n// ── Built-in phrase map ───────────────────────────────────────────────────\n// These are the multi-word expressions handled as compound tokens.\n// Matched by the PhraseMatcher (trie-based) in ExpressionLexer.tryMatchPhrase().\nconst BUILTIN_PHRASES: Record<string, string> = {\n  'to the power of': 'CARET',\n  'power of': 'CARET',\n  'increase by': 'INCREASE_BY',\n  'decrease by': 'DECREASE_BY',\n  'times by': 'TIMES_BY',\n  'multiply by': 'MULTIPLY_BY',\n  'multiplied by': 'MULTIPLY_BY',\n  'divide by': 'DIVIDE_BY',\n};\n\n/**\n * Build a TokenLookup from locale keywords, known units, and built-in phrases.\n *\n * Merge order (later overrides earlier):\n *   1. Locale keywords (priority 0, lowest, can be overridden by providers)\n *   2. Built-in phrases (via locale phraseMap)\n *   3. Known units (checked AFTER keyword lookup fails, via unitNames set)\n *\n * The resulting TokenLookup replaces the internal keyword map, unit set,\n * phrase trie, and phraseStartWords in ExpressionLexer when set via\n * ExpressionLexer.configuredLookup.\n *\n * @param localeCode - The locale code (e.g., \"en\", \"de\"). Defaults to \"en\".\n * @returns A frozen TokenLookup ready for consumption by the lexer.\n */\nexport function buildTokenLookup(localeCode = 'en'): TokenLookup {\n  const locale: ILocale = getLocale(localeCode);\n  const registry = new TokenClassRegistry();\n\n  // Layer 1: Locale keywords (priority 0, lowest)\n  registry.setLocale(locale.keywordMap, BUILTIN_PHRASES);\n\n  // Layer 2: Known units (checked after keyword lookup)\n  registry.setUnits(knownUnits);\n\n  // Build and return the lookup\n  return registry.build();\n}\n"]}