@amritk/yaml - npm Package Compare versions

Comparing version 0.1.0 to 0.1.1

dist/resolve-scalar.d.ts.map

@@ -1,1 +0,1 @@
-{"version":3,"file":"resolve-scalar.d.ts","sourceRoot":"","sources":["../src/resolve-scalar.ts"],"names":[],"mappings":"AAAA;;;;;;;GAOG;AA8BH,8EAA8E;AAC9E,eAAO,MAAM,iBAAiB,SAAU,MAAM,KAAG,MAAM,GAAG,MAAM,GAAG,OAAO,GAAG,IAoD5E,CAAA;AAyDD,oFAAoF;AACpF,eAAO,MAAM,mBAAmB,UAAW,MAAM,KAAG,MAGnD,CAAA;AAED,6FAA6F;AAC7F,eAAO,MAAM,mBAAmB,UAAW,MAAM,KAAG,MAgCnD,CAAA"}
+{"version":3,"file":"resolve-scalar.d.ts","sourceRoot":"","sources":["../src/resolve-scalar.ts"],"names":[],"mappings":"AAAA;;;;;;;GAOG;AA8BH,8EAA8E;AAC9E,eAAO,MAAM,iBAAiB,SAAU,MAAM,KAAG,MAAM,GAAG,MAAM,GAAG,OAAO,GAAG,IAoD5E,CAAA;AA4ED,oFAAoF;AACpF,eAAO,MAAM,mBAAmB,UAAW,MAAM,KAAG,MAGnD,CAAA;AAED,6FAA6F;AAC7F,eAAO,MAAM,mBAAmB,UAAW,MAAM,KAAG,MAgCnD,CAAA"}

dist/resolve-scalar.js

@@ -113,6 +113,16 @@ /**
 /**
- * Folds the line breaks of a multi-line flow scalar: a single break becomes a
- * space, and each blank line becomes one literal newline. Leading and trailing
- * whitespace on continuation lines is trimmed, per the YAML flow folding rules.
+ * Folds the line breaks of a multi-line flow scalar, per the YAML flow folding
+ * rules: a single break between content becomes a space, and a run of blank
+ * lines becomes that many literal newlines.
+ *
+ * Whitespace handling mirrors what the spec keeps as content vs. discards:
+ * - leading whitespace on a continuation line is folding indentation, so it is
+ *   always dropped;
+ * - trailing whitespace is dropped on every line *except the last*, where no
+ *   line break follows so the spaces are literal content;
+ * - a blank-line run that reaches the end of the scalar yields one fewer
+ *   newline, because the break before the closing delimiter is stripped.
  */
+const lstrip = (s) => s.replace(/^[ \t]+/, '');
+const rstrip = (s) => s.replace(/[ \t]+$/, '');
 const foldLines = (text) => {
@@ -122,16 +132,23 @@ const lines = text.split('\n');
         return text;
-    let out = lines[0]?.replace(/[ \t]+$/, '') ?? '';
+    const last = lines.length - 1;
+    let out = rstrip(lines[0] ?? '');
     let i = 1;
-    while (i < lines.length) {
-        const trimmed = (lines[i] ?? '').trim();
-        if (trimmed === '') {
-            // Run of blank lines: each one contributes a newline.
+    while (i <= last) {
+        if ((lines[i] ?? '').trim() === '') {
+            // Run of blank lines.
             let blanks = 0;
-            while (i < lines.length && (lines[i] ?? '').trim() === '') {
+            while (i <= last && (lines[i] ?? '').trim() === '') {
                 blanks++;
                 i++;
             }
-            out += '\n'.repeat(blanks);
-            if (i < lines.length) {
-                out += (lines[i] ?? '').trim();
+            if (i > last) {
+                // Trailing run reaching the closing delimiter: a lone break still folds
+                // to a space; any further blank lines each drop one break, so a run of
+                // `n` contributes `n - 1` newlines.
+                out += blanks === 1 ? ' ' : '\n'.repeat(blanks - 1);
+            }
+            else {
+                // Interior run: each blank line is one newline, then the next content.
+                out += '\n'.repeat(blanks);
+                out += i === last ? lstrip(lines[i] ?? '') : (lines[i] ?? '').trim();
                 i++;
@@ -141,3 +158,5 @@ }
         else {
-            out += ' ' + trimmed;
+            // Single break folds to a space. Keep trailing whitespace only on the
+            // final line, where it is literal content rather than folding padding.
+            out += ' ' + (i === last ? lstrip(lines[i] ?? '') : (lines[i] ?? '').trim());
             i++;
@@ -144,0 +163,0 @@ }

package.json

 {
   "name": "@amritk/yaml",
-  "version": "0.1.0",
+  "version": "0.1.1",
   "description": "A tiny, dependency-free YAML parser with exact source positions. Built for diagnostics: every node maps back to an exact line:column.",
@@ -5,0 +5,0 @@ "module": "./dist/index.js",

README.md

@@ -20,3 +20,3 @@ <div align="center">
 
-`@amritk/yaml` parses YAML into a JavaScript value **and** a lightweight tree where **every node records its exact `[start, end)` source range**. That second part is the whole point: a linter or language server needs to put a squiggle at an exact `line:column`, and most fast YAML parsers throw position information away.
+`@amritk/yaml` parses YAML into a JavaScript value **and** a lightweight tree where **every node records the exact source it came from** — a `start` offset (inclusive) and an `end` offset (exclusive). That second part is the whole point: a linter or language server needs to put a squiggle at an exact `line:column`, and most fast YAML parsers throw position information away.
 
@@ -102,3 +102,3 @@ It is **zero-dependency** and tuned to be **small and fast**. Against the two parsers people reach for on the web:
 | `parseDocument(source, options?)` | Parse to `{ contents, errors, warnings, toJS() }` where every node carries `start`/`end` source offsets. |
-| `nodeAtPath(root, path, closest?)` | Resolve a JSON path to its node (with `range`), optionally falling back to the closest ancestor. |
+| `nodeAtPath(root, path, closest?)` | Resolve a JSON path to its node (carrying `start`/`end`), optionally falling back to the closest ancestor. |
 | `lineCounter(source)` | Build an `offset → { line, col }` mapper (1-based). |
@@ -105,0 +105,0 @@ | `isScalar` / `isMap` / `isSeq` / `isPair` / `isAlias` | Narrowing guards over the node union. |

src/differential.test.ts

@@ -0,1 +1,2 @@
+import { readFileSync } from 'node:fs'
 import { describe, expect, it } from 'vitest'
@@ -8,2 +9,12 @@ import { parse as eemeli } from 'yaml'
 /**
+ * Real-world public specs vendored under `fixtures/` (see `fixtures/README.md`
+ * for provenance). Read from disk so the bytes stay identical to what the
+ * upstream publisher serves. `bench/fixtures.ts` stays synthetic; this is where
+ * we exercise the parser against documents we don't control.
+ */
+const VENDORED = ['digitalocean'] as const
+const readVendored = (name: string): string =>
+  readFileSync(new URL(`../fixtures/${name}.yaml`, import.meta.url), 'utf8')
+
+/**
  * Differential tests against `yaml` (eemeli) — the reference parser the Loupe
@@ -34,2 +45,10 @@ * linter currently uses and the one that, like us, tracks source positions. For
   'q: "https://example.com/a?b=c&d=e"\n',
+  // Multi-line flow scalars — folding edge cases that real specs (GitHub's
+  // OpenAPI) hit: trailing whitespace on the closing line is literal content,
+  // and a blank-line run reaching the close yields one fewer newline.
+  "s: 'first line\n  second line. '\n",
+  's: "first line\n  second line. "\n',
+  "s: 'para one\n  still one\n\n  '\n",
+  's: "para one\n  still one\n\n  "\n',
+  "s: 'a\n\n\n  '\nt: 'a\n  b\n\n  c'\n",
   // Block scalars with chomping.
@@ -59,2 +78,11 @@ 'text: |\n  line one\n  line two\n',
   }
+
+  // Large, real-world public specs we don't control — the documents this
+  // parser actually has to survive in the wild.
+  for (const name of VENDORED) {
+    it(`matches yaml for vendored spec: ${name}`, () => {
+      const source = readVendored(name)
+      expect(ours(source)).toEqual(eemeli(source, { merge: true }))
+    })
+  }
 })

src/resolve-scalar.ts

@@ -115,27 +115,46 @@ /**
 /**
- * Folds the line breaks of a multi-line flow scalar: a single break becomes a
- * space, and each blank line becomes one literal newline. Leading and trailing
- * whitespace on continuation lines is trimmed, per the YAML flow folding rules.
+ * Folds the line breaks of a multi-line flow scalar, per the YAML flow folding
+ * rules: a single break between content becomes a space, and a run of blank
+ * lines becomes that many literal newlines.
+ *
+ * Whitespace handling mirrors what the spec keeps as content vs. discards:
+ * - leading whitespace on a continuation line is folding indentation, so it is
+ *   always dropped;
+ * - trailing whitespace is dropped on every line *except the last*, where no
+ *   line break follows so the spaces are literal content;
+ * - a blank-line run that reaches the end of the scalar yields one fewer
+ *   newline, because the break before the closing delimiter is stripped.
  */
+const lstrip = (s: string): string => s.replace(/^[ \t]+/, '')
+const rstrip = (s: string): string => s.replace(/[ \t]+$/, '')
+
 const foldLines = (text: string): string => {
   const lines = text.split('\n')
   if (lines.length === 1) return text
-  let out = lines[0]?.replace(/[ \t]+$/, '') ?? ''
+  const last = lines.length - 1
+  let out = rstrip(lines[0] ?? '')
   let i = 1
-  while (i < lines.length) {
-    const trimmed = (lines[i] ?? '').trim()
-    if (trimmed === '') {
-      // Run of blank lines: each one contributes a newline.
+  while (i <= last) {
+    if ((lines[i] ?? '').trim() === '') {
+      // Run of blank lines.
       let blanks = 0
-      while (i < lines.length && (lines[i] ?? '').trim() === '') {
+      while (i <= last && (lines[i] ?? '').trim() === '') {
         blanks++
         i++
       }
-      out += '\n'.repeat(blanks)
-      if (i < lines.length) {
-        out += (lines[i] ?? '').trim()
+      if (i > last) {
+        // Trailing run reaching the closing delimiter: a lone break still folds
+        // to a space; any further blank lines each drop one break, so a run of
+        // `n` contributes `n - 1` newlines.
+        out += blanks === 1 ? ' ' : '\n'.repeat(blanks - 1)
+      } else {
+        // Interior run: each blank line is one newline, then the next content.
+        out += '\n'.repeat(blanks)
+        out += i === last ? lstrip(lines[i] ?? '') : (lines[i] ?? '').trim()
         i++
       }
     } else {
-      out += ' ' + trimmed
+      // Single break folds to a space. Keep trailing whitespace only on the
+      // final line, where it is literal content rather than folding padding.
+      out += ' ' + (i === last ? lstrip(lines[i] ?? '') : (lines[i] ?? '').trim())
       i++
@@ -142,0 +161,0 @@ }

No alert changes

Improved metrics

Total package byte prevSize

125490

2.94%

Lines of code

3019

2.13%

No dependency changes