@konvert7/klint 0.28.0 → 0.29.0

package/README.md

@@ -415,6 +415,18 @@ arch:
 
 Set `countDocComments: true` on either rule to count doc-comments toward the limit as well. Both rules run in the TS and Rust engines; Python (`#`) and Swift (`//`, `/* */`) files are enforced by the Rust engine.
 
+Some comments are machinery rather than prose — markers a build or codegen step reads, region pragmas — and counting them makes both limits measure the wrong thing. `ignore` exempts them, on either rule:
+
+```yaml
+arch:
+  maxCommentDensity:
+    - limit: 5
+      in: ["src/**"]
+      ignore: ["@codegen:"]     # literal substring, or "re:" for a regex
+```
+
+An ignored line still counts in the density denominator — it is a physical line like code — and still holds a comment block together, so a marker cannot be sprinkled between comment lines to split an over-tall block. Matching tests the physical source line and therefore only ever applies to lines already lexed as comments.
+
 ### Language Support
 
 klint scans TypeScript/JavaScript, Python, Swift, and Rust. Every architecture rule is enforced from the same `klint.yaml`, but what a rule can *see* depends on what the language actually has.

package/core/arch-comments.ts

@@ -52,22 +52,40 @@ export function commentLineSet(
   return set;
 }
 
-/** Returns the first line at which a run of consecutive comment lines exceeds `limit`. */
+/**
+ * Comment lines whose source text matches an `ignore` pattern. They stay in the
+ * density denominator and keep a comment run contiguous, but count toward neither.
+ */
+export function ignoredCommentLines(
+  content: string,
+  commentLines: Iterable<number>,
+  matchesIgnore: (text: string) => boolean
+): Set<number> {
+  const sourceLines = content.split("\n");
+  const ignored = new Set<number>();
+  for (const line of commentLines) {
+    if (matchesIgnore(sourceLines[line - 1] ?? "")) ignored.add(line);
+  }
+  return ignored;
+}
+
+/**
+ * Returns the first line at which a run of consecutive comment lines exceeds
+ * `limit`. An ignored line neither counts toward the height nor breaks the run.
+ */
 export function firstCommentBlockOverrun(
   sortedLines: number[],
-  limit: number
+  limit: number,
+  ignored?: ReadonlySet<number>
 ): number | undefined {
-  let runStart = sortedLines[0];
-  let prev = sortedLines[0];
-  for (let i = 1; i <= sortedLines.length; i++) {
-    const line = sortedLines[i];
-    if (line === prev + 1) {
-      prev = line;
-      continue;
-    }
-    if (prev - runStart + 1 > limit) return runStart + limit;
-    runStart = line;
+  let counted = 0;
+  let prev: number | undefined;
+  for (const line of sortedLines) {
+    if (prev !== undefined && line !== prev + 1) counted = 0;
     prev = line;
+    if (ignored?.has(line)) continue;
+    counted += 1;
+    if (counted > limit) return line;
   }
   return undefined;
 }

package/core/arch.ts

@@ -1,6 +1,10 @@
 import { dirname, isAbsolute, resolve } from "node:path";
 import ts from "typescript";
-import { commentLineSet, firstCommentBlockOverrun } from "./arch-comments";
+import {
+  commentLineSet,
+  firstCommentBlockOverrun,
+  ignoredCommentLines,
+} from "./arch-comments";
 import { walkAst } from "./ast";
 import { relativeSlashPath, toSlashPath } from "./paths";
 import type { ArchConfig, Severity, Violation } from "./types";
@@ -309,6 +313,7 @@ export function runArchRules(
 
   for (const rule of arch.maxCommentDensity ?? []) {
     const severity: Severity = rule.severity ?? "error";
+    const ignoreMatcher = buildIgnoreMatcher(rule.ignore);
     const inFiles = resolveLayerFiles(rule.in, layers, root, allFiles);
     for (const file of inFiles) {
       const content = fileContents.get(file);
@@ -316,7 +321,11 @@ export function runArchRules(
       const total = countLines(content);
       if (total === 0) continue;
       const commentLines = commentLineSet(file, content, rule.countDocComments ?? false);
-      const density = (commentLines.size / total) * 100;
+      const ignored = ignoreMatcher
+        ? ignoredCommentLines(content, commentLines, ignoreMatcher)
+        : undefined;
+      const counted = commentLines.size - (ignored?.size ?? 0);
+      const density = (counted / total) * 100;
       if (density > rule.limit) {
         violations.push({
           file: relativeSlashPath(root, file),
@@ -333,6 +342,7 @@ export function runArchRules(
 
   for (const rule of arch.maxCommentBlock ?? []) {
     const severity: Severity = rule.severity ?? "error";
+    const ignoreMatcher = buildIgnoreMatcher(rule.ignore);
     const inFiles = resolveLayerFiles(rule.in, layers, root, allFiles);
     for (const file of inFiles) {
       const content = fileContents.get(file);
@@ -340,7 +350,10 @@ export function runArchRules(
       const commentLines = [
         ...commentLineSet(file, content, rule.countDocComments ?? false),
       ].sort((a, b) => a - b);
-      const overrun = firstCommentBlockOverrun(commentLines, rule.limit);
+      const ignored = ignoreMatcher
+        ? ignoredCommentLines(content, commentLines, ignoreMatcher)
+        : undefined;
+      const overrun = firstCommentBlockOverrun(commentLines, rule.limit, ignored);
       if (overrun !== undefined) {
         violations.push({
           file: relativeSlashPath(root, file),
@@ -402,6 +415,15 @@ function scanJsxElements(
 
 const REGEX_PREFIX = "re:";
 
+function buildIgnoreMatcher(
+  ignore: string | string[] | undefined
+): ((line: string) => boolean) | undefined {
+  const patterns = toArray(ignore ?? []);
+  if (patterns.length === 0) return undefined;
+  const matchers = patterns.map(buildLineMatcher);
+  return (line) => matchers.some((matches) => matches(line));
+}
+
 function buildLineMatcher(pattern: string): (line: string) => boolean {
   if (!pattern.startsWith(REGEX_PREFIX)) {
     return (line) => line.includes(pattern);

package/core/config.schema.ts

@@ -247,6 +247,10 @@ const CountDocCommentsSchema = z
     "Count doc-comments toward the limit. Doc-comments are `/** … */` JSDoc blocks (TS/JS) and docstrings (Python); ordinary `//`, `/* */`, and `#` comments are always counted. Defaults to false, so documentation is exempt and only explanatory/inline comments are capped."
   );
 
+const CommentIgnoreSchema = StringOrStringArray.optional().describe(
+  "Comment lines matching these patterns are structural, not prose — marker comments a build or codegen step reads, for example — and count toward neither limit. A literal substring by default; prefix with `re:` for a regex. Ignored lines still count in the density denominator (they are physical lines) and still hold a comment block together, so a marker cannot be used to split an over-tall block."
+);
+
 const ArchMaxCommentDensityRuleSchema = z
   .object({
     limit: z
@@ -257,6 +261,7 @@ const ArchMaxCommentDensityRuleSchema = z
         "Maximum percentage of a file's physical lines that may be comment lines. Denominator is total physical lines (code + comments + blanks), the same count `maxLines` uses. A file above this ratio is a violation."
       ),
     countDocComments: CountDocCommentsSchema,
+    ignore: CommentIgnoreSchema,
     in: StringOrStringArray.describe(
       "Layer name(s) or glob(s) the limit applies to. Files outside the scope are not checked."
     ),
@@ -289,6 +294,7 @@ const ArchMaxCommentBlockRuleSchema = z
         "Maximum number of consecutive comment lines allowed in a single block. A run of more comment lines than this — e.g. a tall comment stacked above a statement — is a violation, reported at the first offending line."
       ),
     countDocComments: CountDocCommentsSchema,
+    ignore: CommentIgnoreSchema,
     in: StringOrStringArray.describe(
       "Layer name(s) or glob(s) the limit applies to. Files outside the scope are not checked."
     ),

package/core/types.ts

@@ -110,6 +110,8 @@ interface ArchCommentRuleBase {
   in: string | string[];
   /** Count doc-comments (`/** *​/` JSDoc, Python docstrings) toward the limit. Defaults to false. */
   countDocComments?: boolean;
+  /** Comment lines matching these patterns count toward neither limit. Literal substring, or `re:` for a regex. */
+  ignore?: string | string[];
   message?: string;
   severity?: Exclude<Severity, "off">;
 }

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@konvert7/klint",
-  "version": "0.28.0",
+  "version": "0.29.0",
   "description": "Architecture-as-Code linter for TypeScript. Enforce layer boundaries, import rules, and singleton patterns in YAML.",
   "type": "module",
   "bin": {
@@ -65,10 +65,10 @@
     "zod": "^4.4.3"
   },
   "optionalDependencies": {
-    "@konvert7/klint-darwin-arm64": "0.28.0",
-    "@konvert7/klint-darwin-x64": "0.28.0",
-    "@konvert7/klint-linux-x64": "0.28.0",
-    "@konvert7/klint-win32-x64": "0.28.0"
+    "@konvert7/klint-darwin-arm64": "0.29.0",
+    "@konvert7/klint-darwin-x64": "0.29.0",
+    "@konvert7/klint-linux-x64": "0.29.0",
+    "@konvert7/klint-win32-x64": "0.29.0"
   },
   "devDependencies": {
     "@biomejs/biome": "2.4.7",

package/skill/klint-rules/SKILL.md

@@ -39,6 +39,8 @@ Wire it into `klint.yaml` for editor autocomplete and validation:
    | Raw HTML/JSX element may only appear in one primitive file | `arch.singleton` + `jsx-element` |
    | A value-shaped pattern (regex) must never appear in a scoped layer | `arch.forbidden` + `pattern: "re:…"` |
    | No file in a layer may exceed N lines | `arch.maxLines` + `limit` |
+   | No file may be more than N% comments | `arch.maxCommentDensity` + `limit` |
+   | No comment block may stack more than N lines | `arch.maxCommentBlock` + `limit` |
 
 4. **Read `klint.yaml`** — check existing `arch.layers` and rules before adding anything. Add a new named layer to `arch.layers` if the file group doesn't exist yet.
 
@@ -169,10 +171,65 @@ arch:
 
 Required: `limit` + `in`. The count is **total physical lines** — blanks and comments included, not code lines — and a file over the limit is flagged at line `limit + 1`. Enforced identically in the TS and Rust engines, across every language klint scans.
 
+### Comment budget — cap explanatory prose
+
+Two ceilings on how much of a file may be comments. Use them to push intent into names and small
+functions rather than paragraphs that drift from the code they describe.
+
+```yaml
+arch:
+  # maxCommentDensity: what share of a file may be comment lines
+  maxCommentDensity:
+    - limit: 5                 # percent; required
+      in: ["src/**"]           # string OR array; supports ! negation
+      countDocComments: false  # optional — default false, so JSDoc/docstrings are exempt
+      message: "Encode intent in names and small functions, not prose"
+      severity: error          # optional — default error; use warn to record without blocking
+
+  # maxCommentBlock: how tall a single stack of comment lines may get
+  maxCommentBlock:
+    - limit: 2                 # consecutive comment lines; required
+      in: ["src/**"]
+      ignore: ["@codegen:"]    # optional — structural markers, not prose
+      message: "Extract a well-named function instead of stacking comment lines"
+```
+
+Required: `limit` + `in`. Density is a **percentage of total physical lines** (code + comments +
+blanks — the same denominator `maxLines` uses). Block height counts **consecutive** comment lines and
+is reported at the first offending line. Both exempt doc-comments by default — `/** … */` JSDoc and
+Python docstrings — so documentation is not penalised; ordinary `//`, `/* */`, and `#` comments always
+count. Set `countDocComments: true` to include them.
+
+The two are complementary: density catches a file that is 30% commented in scattered one-liners,
+block catches a ten-line essay above one function in an otherwise sparse file. Neither alone catches both.
+
+**`ignore` — comments that are machinery, not prose.** Some comments are load-bearing input to a
+build or codegen step (`// @codegen:region-start`, feature-strip markers, pragmas). They are comments to the
+lexer but structure to the project, and counting them makes both limits measure the wrong thing:
+
+```yaml
+arch:
+  maxCommentDensity:
+    - limit: 5
+      in: ["src/**"]
+      ignore: ["@codegen:"]         # literal substring, or "re:" for a regex
+```
+
+Two rules govern how an ignored line behaves, and both matter:
+
+- **It stays in the density denominator.** A marker is a physical line like code, so a 20-line file
+  with 4 markers and 1 real comment measures 1/20, not 1/16.
+- **It does not break a comment block.** `// a` / `// @marker` / `// b` / `// c` is a run of three
+  counted lines, not two runs. Otherwise sprinkling a marker every other line would defeat the cap.
+
+`ignore` tests the **physical source line**, so it only ever applies to lines already identified as
+comments — code containing the same text is untouched. Enforced identically in both engines.
+
 ## Pitfalls
 
 - `pattern` in `singleton` and `forbidden` is a **literal substring** by default — it will not catch `process.env["KEY"]` (bracket notation). Either grep for both forms, or use a `re:` regex prefix to match them in one rule. Caveat: a literal pattern that itself begins with `re:` cannot be expressed (it is always read as a regex), and regexes must stay in the common JS/RE2 subset (no lookaround or backreferences) so both engines agree.
 - `maxLines.limit` counts **total physical lines** (blanks and comments included), not code lines; a trailing newline does not add a line, and both engines count identically.
+- `maxCommentDensity.limit` is a **percentage, not a line count** — `limit: 5` means 5% of the file, so a 40-line file is over budget at its third comment line. Measure before picking a number: land it as `severity: warn` first, read the reported densities, then tighten to `error`.
 - `jsx-element` matches **intrinsic** (lowercase HTML) tags only — `button`, `input`, `label`. It does not match custom React components like `<Button>`; to restrict those, use `arch.imports` against the component's module instead.
 - A `forbidden`/`singleton` stanza takes either `pattern` or `jsx-element`, never both — split into two stanzas if you need both.
 - `only` in `singleton` is relative to the project root — use forward slashes on all platforms.