@mulmoclaude/markdown-utils 2.0.0 → 2.2.0

package/dist/markdown/mathExtension.js

@@ -32,8 +32,34 @@
 //      and must not be a backslash.
 //   4. The character AFTER the closing `$` must not be an ASCII digit
 //      (`$5-$10`).
-//   5. The body must be non-empty, single-line, and must not consist
-//      only of digits and separators (`$1,000$`).
+//   5. The body must be non-empty, single-line, and must be something a
+//      formula could be ABOUT: a number written the way money is
+//      (`$1,000$`, `$1.000,50$` — a price written twice) or separators
+//      with no digits at all (`$+$`) are not.
+//
+// Rule 5 used to reject EVERY digits-and-separators body, and that was
+// too wide: `1秒を $10000$ 個のステップに割る` and `答えは $1$` are the
+// ordinary way to write a number in a maths article, and both came out
+// as a literal `$10000$` / `$1$` sitting in the prose. The signature of
+// a price is its SHAPE — digits in threes, or more than one separator —
+// and the two shapes that actually
+// appear in currency prose are already dead: `$100 と $200` by rule 3
+// (whitespace before the close) and `$5-$10` by rule 4 (a digit after
+// it). What stays admitted is a body like `$5$`, which a person quoting
+// a price does not write: they write `$5`, and it is the DOUBLED
+// delimiter that makes it maths.
+//
+// The SHAPE is matched rather than a particular separator, because a
+// comma groups thousands in English and marks the decimal in most of
+// Europe, and a dot does the opposite (codex, #2985). `$1,5$` is one and
+// a half and typesets; `$1.000,50$` carries both a grouping AND a
+// decimal mark, which no one computes with, and does not.
+//
+// A number with ONE separator typesets whatever follows it. Reading
+// three digits after it as money cost every three-decimal constant a
+// maths article writes (`$3.141$`, `$1.414$`, `$2.718$`) and bought only
+// `$1,500$` — a shape currency prose never reaches, because rules 1-4
+// kill every realistic way of writing a price first (#2991).
 //
 // Rules 2-5 are enforced in the tokenizer, where the whole match is in
 // hand. Rule 1 needs the character BEFORE the match, which a marked
@@ -49,9 +75,54 @@
 import { escapeHtml } from "@mulmoclaude/common";
 const ASCII_ALNUM = /[A-Za-z0-9]/;
 const ASCII_DIGIT = /\d/;
-/** Digits, separators and currency-ish punctuation only — `$1,000$`
- *  is a price range, not an equation. */
-const NUMERIC_ONLY = /^[\s\d.,:;%+-]*$/;
+const DIGITS = /^\d+$/;
+/** Every character that separates the digit runs of a written number,
+ *  in any locale: `1,000.50`, `1.000,50`, `1 000,50`. Which one groups
+ *  and which one marks the decimal is exactly what cannot be known, so
+ *  none of them is read as one or the other. */
+const NUMBER_SEPARATORS = /[.,\s]/;
+/** A bare number written the way MONEY is written rather than the way a
+ *  quantity is — `1,000`, `1.000`, `1 000`, `1.000,50`, `12,345,678`.
+ *
+ *  Read from the SHAPE, not from which separator appeared: a comma
+ *  groups thousands in English and marks the decimal in most of Europe,
+ *  and a dot does the opposite, so a rule that names one of them fails
+ *  half the world's authors either way (codex, #2985).
+ *
+ *  Two separators or more is money: `1.000,50` groups AND marks a
+ *  decimal, which is a formatted amount rather than a quantity anyone
+ *  computes with.
+ *
+ *  ONE separator is not, whatever follows it. That branch used to read
+ *  "exactly three digits after it" as money, and the cost was every
+ *  three-decimal constant a maths article writes — `3.141`, `1.414`,
+ *  `2.718`, `1.618`, `0.577`, `6.022`, `9.807` all came out as literal
+ *  `$…$` in the prose (#2991). What it bought was `$1,500$`, and that
+ *  turns out to be a shape currency prose never reaches: every realistic
+ *  form dies earlier, at rules 1-4 —
+ *
+ *      `$1,000 と $2,000`      rule 3, whitespace before the close
+ *      `合計 $1,000から$500`    rule 3
+ *      `$1,000-$2,000`        rule 4, a digit after the close
+ *      `US$1,000`             rule 1, alphanumeric before the open
+ *      `価格は $1,000 です`      no closing `$` at all
+ *
+ *  — leaving only `$1,000$`, both delimiters written out. That is the
+ *  same shape this file already reads as maths for `$5$`: a person
+ *  quoting a price writes `$1,000`, and it is the DOUBLED delimiter that
+ *  makes it a formula. Rule 5 was applying one reading to `$5$` and the
+ *  opposite to `$1,500$`; now it applies the same one to both. */
+function isMoneyShaped(body) {
+    const runs = body.split(NUMBER_SEPARATORS);
+    // Anything that is not digits-and-separators is not a written number at
+    // all — `x=1` and `\pi` land here and are maths by this rule.
+    if (!runs.every((run) => DIGITS.test(run)))
+        return false;
+    return runs.length > 2;
+}
+/** The same set with the digits removed: a body of punctuation has
+ *  nothing to typeset. */
+const SEPARATORS_ONLY = /^[\s.,:;%+-]*$/;
 /** Index of the first `$` in `src` that could legally open math, or
  *  `undefined` when there is none. Marked uses this to cut the
  *  preceding text token, so returning a position is what gets the
@@ -86,8 +157,11 @@ export function isPlausibleInlineMath(body, after) {
     // Rule 4: `$5-$10`.
     if (ASCII_DIGIT.test(after))
         return false;
-    // Rule 5.
-    if (NUMERIC_ONLY.test(body))
+    // Rule 5. The shape of money is what says "price"; a plain number is
+    // just a number, and a number in a maths article is maths.
+    if (isMoneyShaped(body))
+        return false;
+    if (SEPARATORS_ONLY.test(body))
         return false;
     return true;
 }

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@mulmoclaude/markdown-utils",
-  "version": "2.0.0",
+  "version": "2.2.0",
   "description": "Browser-safe markdown / image rendering utilities shared by the MulmoClaude host and the markdown plugin",
   "type": "module",
   "main": "./dist/index.js",
@@ -35,7 +35,7 @@
   "dependencies": {
     "@mulmoclaude/common": "^1.2.0",
     "dompurify": "^3.4.13",
-    "js-yaml": "^5.2.3",
+    "js-yaml": "^5.4.1",
     "marked": "^18.0.11"
   },
   "peerDependencies": {
@@ -47,7 +47,7 @@
     "mathjax-full": "^3.2.2",
     "mermaid": "^11.17.2",
     "typescript": "^6.0.3",
-    "vue": "^3.5.41"
+    "vue": "^3.5.42"
   },
   "homepage": "https://github.com/receptron/mulmoclaude/tree/main/packages/markdown-utils#readme",
   "repository": {