@ai-matrx/content-ir 0.16.2 → 0.18.0

package/dist/source.d.cts

@@ -300,21 +300,48 @@ declare function toCodePointOffset(text: string, utf16: number): number;
 declare function toUtf16Offset(text: string, codePoint: number): number;
 
 /**
- * THE single-dollar math rule — the one predicate that decides whether
- * `$…$` is inline math or currency.
+ * THE single-dollar math rule — the one predicate that decides whether a
+ * CLOSED `$…$` is inline math or currency, on screen, in every export and in
+ * the editor's islands (the renderer's normalizer, the source tokenizer and
+ * print all ask it).
  *
- * Moved here from matrx-frontend `components/markdown-core/math-normalizer.ts`
- * (2026-09-23) so the renderer's normalizer and the source tokenizer can never
- * disagree about which bytes are math: the normalizer imports this function.
- *
- * Pandoc's rule — the opening `$` is not followed by whitespace, the closing
- * `$` is not preceded by whitespace and not followed by a digit — plus a math
- * signal: a TeX command or operator (`\alpha`, `^`, `_`, `{`, `=`, …), a
- * lone variable (`$x$`, `$n$`, `$x'$`), or a single-letter math shape
- * (`$f(g(x))$`, `$u v$`, `$f'(x)$` — see `isMathShape`). Content that opens with a digit needs
- * a real TeX signal, so a price range can never read as math.
+ * Pandoc's `tex_math_dollars` delimiters (what remark-math approximates, and
+ * how ChatGPT and Claude render): the opening `$` is followed by a
+ * non-space, the closing `$` is preceded by a non-space and not followed by a
+ * digit — so `$5 and $10`, `$5-$10`, `$5/$10` stay currency while `$FC$`,
+ * `$AB$`, `$5$` and `$(0, 0)$` are math. On top of that, content that reads as
+ * something other than a formula is refused — each guard is a false positive
+ * measured in stored answers (matrx-frontend
+ * `scripts/measure-single-dollar-rule.ts`, 2026-09-25: over 85k assistant
+ * messages and 4k notes the widening adds 128 spans, every one a formula
+ * (`$FC$`, `$(0, 0)$`, `$3x + 6 = 15$`), and drops 22 the old rule drew as
+ * math that were code, template literals, a printf or broken TeX):
+ *   - an unescaped `%` — TeX's comment character, and a percentage
+ *     (`$\ge$20% owners**…**$`, a printf `%1$s`);
+ *   - a closing bracket or `/` first, or an opening bracket last
+ *     (`[$X] plus [$Y]`, `($/SF/Yr) | High-Cost Market ($`, a regex
+ *     `/…/, '').replace(/…$`), or a straight quote at either end (`["$", "$"]`
+ *     in data);
+ *   - `:` `;` `,` `/` last — a separator between two `$` words
+ *     (`$HOME/.local/bin:$PATH`, `$AAPL/$TSLA`);
+ *   - a number glued to a following letter (`$10$N9qo…`, a bcrypt cost);
+ *   - a placeholder (`$...$`);
+ *   - prose, unless a TeX command or operator says otherwise: two plain
+ *     words (`$Tipping (fee)$`), a contraction (`$I'm$`, never `$X'Y$`), an
+ *     abbreviation (`$e.g. (x)$`).
  */
 declare function isSingleDollarMath(content: string, after: string | undefined): boolean;
+/**
+ * Whether an UNCLOSED `$…` already reads as a formula — for a stream that
+ * must decide before the closing `$` arrives (matrx-frontend `stream-heal`
+ * holds such a tail back instead of printing raw TeX). Stricter than
+ * {@link isSingleDollarMath} on purpose: with no closer there are no Pandoc
+ * delimiter facts yet, so it needs a math SIGNAL — a TeX command or operator
+ * (`\alpha`, `^`, `_`, `{`, `=`, …), a lone variable (`x`, `x'`), or a
+ * single-letter math shape (`f(g(x))`, `u v`). Content that opens with a digit
+ * needs a real TeX command, so `$35 per stop` keeps showing while it streams.
+ */
+declare function looksLikeOpenInlineMath(partial: string): boolean;
 /**
  * Where a single-dollar math span opened at `open` (a lone `$`) closes, or -1.
  * Same scan the normalizer uses: same line, backslash-escapes skipped, the
@@ -420,6 +447,20 @@ declare function parseFenceOpener(line: string): {
 } | null;
 declare function classifyInnerFenceLine(trimmed: string, openTicks: number, nests: boolean, nestedDepth: number): InnerFenceLine;
 
+interface CodeRange {
+    /** UTF-16 offset of the first character (the fence line start, or the opening backtick run). */
+    start: number;
+    /** Offset just past the closing fence line's content / closing backtick run (the fence's last line when unclosed). */
+    end: number;
+    kind: "fence" | "span";
+}
+/**
+ * Every code range in the text — fences (by the renderer's rule above) and
+ * inline code spans — sorted by start. Linear: one pass over lines, one over
+ * each paragraph's backtick runs.
+ */
+declare function findCodeRanges(text: string): CodeRange[];
+
 /**
  * `$$...$$`, `$$…$$`, `$$ · $$`: a PLACEHOLDER — prose describing the syntax,
  * never a formula. The pair is consumed whole as literal text (neither `$$`
@@ -433,7 +474,7 @@ declare function isMathPlaceholder(inner: string): boolean;
  * pairing with the next paragraph's formula.
  */
 declare function looksLikeDisplayMath(inner: string, flow?: boolean): boolean;
-/** Map of `$$` opener offset → closer offset for every pair the renderer shows as math. */
+
 /**
  * STREAMING: the last `$$` token that is still unpaired AND could still pair
  * with a `$$` not yet arrived — its span so far is short enough, carries no
@@ -442,6 +483,7 @@ declare function looksLikeDisplayMath(inner: string, flow?: boolean): boolean;
  * streaming tokenizer treats it as provisional. -1 when there is none.
  */
 declare function pendingDisplayMathOpener(text: string): number;
+/** Map of `$$` opener offset → closer offset for every pair the renderer shows as math. */
 declare function pairDisplayMath(text: string): Map<number, number>;
 
-export { type BlockIslandType, CALLOUT_NAMES, type CodePointIndex, type DisturbedIsland, FENCE_OPENER_WHITESPACE, FENCE_WHITESPACE, type InlineIslandType, type InnerFenceLine, type IslandMeta, type IslandType, type MapOptions, type MappedPosition, type MappedRange, NESTING_FENCE_LANGUAGES, type ReadXmlTag, type SourceBlock, type SourceBlockKind, type SourceChange, type SourceEdit, type SourceIsland, type SourceSpan, SourceSpliceError, type SourceSpliceErrorCode, type SpliceIntegrity, type SpliceOptions, type SpliceResult, type TokenizeOptions, XmlContainerTracker, blockAt, blockEdit, blockRangeEdit, buildCodePointIndex, classifyInnerFenceLine, fenceNestsInnerFences, isMathPlaceholder, isSingleDollarMath, islandEdit, joinSource, listIslands, looksLikeDisplayMath, mapPosition, mapRange, pairDisplayMath, parseFenceOpener, pendingDisplayMathOpener, readXmlTag, singleDollarMathEnd, spliceSave, splitsSurrogatePair, toCodePointOffset, toUtf16Offset, tokenizeSource, trimFenceLine };
+export { type BlockIslandType, CALLOUT_NAMES, type CodePointIndex, type CodeRange, type DisturbedIsland, FENCE_OPENER_WHITESPACE, FENCE_WHITESPACE, type InlineIslandType, type InnerFenceLine, type IslandMeta, type IslandType, type MapOptions, type MappedPosition, type MappedRange, NESTING_FENCE_LANGUAGES, type ReadXmlTag, type SourceBlock, type SourceBlockKind, type SourceChange, type SourceEdit, type SourceIsland, type SourceSpan, SourceSpliceError, type SourceSpliceErrorCode, type SpliceIntegrity, type SpliceOptions, type SpliceResult, type TokenizeOptions, XmlContainerTracker, blockAt, blockEdit, blockRangeEdit, buildCodePointIndex, classifyInnerFenceLine, fenceNestsInnerFences, findCodeRanges, isMathPlaceholder, isSingleDollarMath, islandEdit, joinSource, listIslands, looksLikeDisplayMath, looksLikeOpenInlineMath, mapPosition, mapRange, pairDisplayMath, parseFenceOpener, pendingDisplayMathOpener, readXmlTag, singleDollarMathEnd, spliceSave, splitsSurrogatePair, toCodePointOffset, toUtf16Offset, tokenizeSource, trimFenceLine };

package/dist/source.d.ts

@@ -300,21 +300,48 @@ declare function toCodePointOffset(text: string, utf16: number): number;
 declare function toUtf16Offset(text: string, codePoint: number): number;
 
 /**
- * THE single-dollar math rule — the one predicate that decides whether
- * `$…$` is inline math or currency.
+ * THE single-dollar math rule — the one predicate that decides whether a
+ * CLOSED `$…$` is inline math or currency, on screen, in every export and in
+ * the editor's islands (the renderer's normalizer, the source tokenizer and
+ * print all ask it).
  *
- * Moved here from matrx-frontend `components/markdown-core/math-normalizer.ts`
- * (2026-09-23) so the renderer's normalizer and the source tokenizer can never
- * disagree about which bytes are math: the normalizer imports this function.
- *
- * Pandoc's rule — the opening `$` is not followed by whitespace, the closing
- * `$` is not preceded by whitespace and not followed by a digit — plus a math
- * signal: a TeX command or operator (`\alpha`, `^`, `_`, `{`, `=`, …), a
- * lone variable (`$x$`, `$n$`, `$x'$`), or a single-letter math shape
- * (`$f(g(x))$`, `$u v$`, `$f'(x)$` — see `isMathShape`). Content that opens with a digit needs
- * a real TeX signal, so a price range can never read as math.
+ * Pandoc's `tex_math_dollars` delimiters (what remark-math approximates, and
+ * how ChatGPT and Claude render): the opening `$` is followed by a
+ * non-space, the closing `$` is preceded by a non-space and not followed by a
+ * digit — so `$5 and $10`, `$5-$10`, `$5/$10` stay currency while `$FC$`,
+ * `$AB$`, `$5$` and `$(0, 0)$` are math. On top of that, content that reads as
+ * something other than a formula is refused — each guard is a false positive
+ * measured in stored answers (matrx-frontend
+ * `scripts/measure-single-dollar-rule.ts`, 2026-09-25: over 85k assistant
+ * messages and 4k notes the widening adds 128 spans, every one a formula
+ * (`$FC$`, `$(0, 0)$`, `$3x + 6 = 15$`), and drops 22 the old rule drew as
+ * math that were code, template literals, a printf or broken TeX):
+ *   - an unescaped `%` — TeX's comment character, and a percentage
+ *     (`$\ge$20% owners**…**$`, a printf `%1$s`);
+ *   - a closing bracket or `/` first, or an opening bracket last
+ *     (`[$X] plus [$Y]`, `($/SF/Yr) | High-Cost Market ($`, a regex
+ *     `/…/, '').replace(/…$`), or a straight quote at either end (`["$", "$"]`
+ *     in data);
+ *   - `:` `;` `,` `/` last — a separator between two `$` words
+ *     (`$HOME/.local/bin:$PATH`, `$AAPL/$TSLA`);
+ *   - a number glued to a following letter (`$10$N9qo…`, a bcrypt cost);
+ *   - a placeholder (`$...$`);
+ *   - prose, unless a TeX command or operator says otherwise: two plain
+ *     words (`$Tipping (fee)$`), a contraction (`$I'm$`, never `$X'Y$`), an
+ *     abbreviation (`$e.g. (x)$`).
  */
 declare function isSingleDollarMath(content: string, after: string | undefined): boolean;
+/**
+ * Whether an UNCLOSED `$…` already reads as a formula — for a stream that
+ * must decide before the closing `$` arrives (matrx-frontend `stream-heal`
+ * holds such a tail back instead of printing raw TeX). Stricter than
+ * {@link isSingleDollarMath} on purpose: with no closer there are no Pandoc
+ * delimiter facts yet, so it needs a math SIGNAL — a TeX command or operator
+ * (`\alpha`, `^`, `_`, `{`, `=`, …), a lone variable (`x`, `x'`), or a
+ * single-letter math shape (`f(g(x))`, `u v`). Content that opens with a digit
+ * needs a real TeX command, so `$35 per stop` keeps showing while it streams.
+ */
+declare function looksLikeOpenInlineMath(partial: string): boolean;
 /**
  * Where a single-dollar math span opened at `open` (a lone `$`) closes, or -1.
  * Same scan the normalizer uses: same line, backslash-escapes skipped, the
@@ -420,6 +447,20 @@ declare function parseFenceOpener(line: string): {
 } | null;
 declare function classifyInnerFenceLine(trimmed: string, openTicks: number, nests: boolean, nestedDepth: number): InnerFenceLine;
 
+interface CodeRange {
+    /** UTF-16 offset of the first character (the fence line start, or the opening backtick run). */
+    start: number;
+    /** Offset just past the closing fence line's content / closing backtick run (the fence's last line when unclosed). */
+    end: number;
+    kind: "fence" | "span";
+}
+/**
+ * Every code range in the text — fences (by the renderer's rule above) and
+ * inline code spans — sorted by start. Linear: one pass over lines, one over
+ * each paragraph's backtick runs.
+ */
+declare function findCodeRanges(text: string): CodeRange[];
+
 /**
  * `$$...$$`, `$$…$$`, `$$ · $$`: a PLACEHOLDER — prose describing the syntax,
  * never a formula. The pair is consumed whole as literal text (neither `$$`
@@ -433,7 +474,7 @@ declare function isMathPlaceholder(inner: string): boolean;
  * pairing with the next paragraph's formula.
  */
 declare function looksLikeDisplayMath(inner: string, flow?: boolean): boolean;
-/** Map of `$$` opener offset → closer offset for every pair the renderer shows as math. */
+
 /**
  * STREAMING: the last `$$` token that is still unpaired AND could still pair
  * with a `$$` not yet arrived — its span so far is short enough, carries no
@@ -442,6 +483,7 @@ declare function looksLikeDisplayMath(inner: string, flow?: boolean): boolean;
  * streaming tokenizer treats it as provisional. -1 when there is none.
  */
 declare function pendingDisplayMathOpener(text: string): number;
+/** Map of `$$` opener offset → closer offset for every pair the renderer shows as math. */
 declare function pairDisplayMath(text: string): Map<number, number>;
 
-export { type BlockIslandType, CALLOUT_NAMES, type CodePointIndex, type DisturbedIsland, FENCE_OPENER_WHITESPACE, FENCE_WHITESPACE, type InlineIslandType, type InnerFenceLine, type IslandMeta, type IslandType, type MapOptions, type MappedPosition, type MappedRange, NESTING_FENCE_LANGUAGES, type ReadXmlTag, type SourceBlock, type SourceBlockKind, type SourceChange, type SourceEdit, type SourceIsland, type SourceSpan, SourceSpliceError, type SourceSpliceErrorCode, type SpliceIntegrity, type SpliceOptions, type SpliceResult, type TokenizeOptions, XmlContainerTracker, blockAt, blockEdit, blockRangeEdit, buildCodePointIndex, classifyInnerFenceLine, fenceNestsInnerFences, isMathPlaceholder, isSingleDollarMath, islandEdit, joinSource, listIslands, looksLikeDisplayMath, mapPosition, mapRange, pairDisplayMath, parseFenceOpener, pendingDisplayMathOpener, readXmlTag, singleDollarMathEnd, spliceSave, splitsSurrogatePair, toCodePointOffset, toUtf16Offset, tokenizeSource, trimFenceLine };
+export { type BlockIslandType, CALLOUT_NAMES, type CodePointIndex, type CodeRange, type DisturbedIsland, FENCE_OPENER_WHITESPACE, FENCE_WHITESPACE, type InlineIslandType, type InnerFenceLine, type IslandMeta, type IslandType, type MapOptions, type MappedPosition, type MappedRange, NESTING_FENCE_LANGUAGES, type ReadXmlTag, type SourceBlock, type SourceBlockKind, type SourceChange, type SourceEdit, type SourceIsland, type SourceSpan, SourceSpliceError, type SourceSpliceErrorCode, type SpliceIntegrity, type SpliceOptions, type SpliceResult, type TokenizeOptions, XmlContainerTracker, blockAt, blockEdit, blockRangeEdit, buildCodePointIndex, classifyInnerFenceLine, fenceNestsInnerFences, findCodeRanges, isMathPlaceholder, isSingleDollarMath, islandEdit, joinSource, listIslands, looksLikeDisplayMath, looksLikeOpenInlineMath, mapPosition, mapRange, pairDisplayMath, parseFenceOpener, pendingDisplayMathOpener, readXmlTag, singleDollarMathEnd, spliceSave, splitsSurrogatePair, toCodePointOffset, toUtf16Offset, tokenizeSource, trimFenceLine };