Use CR escape in string_cr_escape reject fixture

AGENTS.md

@@ -14,6 +14,8 @@
 - Treat `docs/specification.md` as the authoritative source of behavior; examples may be outdated or incorrect.
 - If the spec is unclear or conflicts with code/tests, ask before changing behavior.
 - Do not implement "just make it work" fallbacks that alter semantics to satisfy examples.
+- Diagnostics must not store dynamic strings (e.g., slices to parsed source). Keep diagnostic payloads POD/small and avoid holding arena-backed text.
+- Do not hide crashes by removing safety checks or switching off DebugAllocator; fix the root cause instead. A signal 6 from DebugAllocator indicates memory corruption or a similar misuse.
 
 ## Zig Programming Style
 
@@ -25,4 +27,4 @@
 - If you add a `hdoc` file to `test/snapshot`, also:
   - Generate the corresponding html and yaml file
   - Add the file inside build.zig to the snapshot_files global
-- If you change behaviour, the snapshot tests will fail. Validate the failure against your expectations and see if you broke something unexpected.
+- If you change behaviour, the snapshot tests will fail. Validate the failure against your expectations and see if you broke something unexpected.

SPEC_TODO.md

@@ -1,2 +1,6 @@
 # Spec compliance TODOs
 
+- Title/header interplay lacks the required comparison.  
+  - Expect: When both `hdoc(title=...)` and `title { ... }` are present, their plaintext forms are compared and a redundancy hint is emitted if they match (§8.1).  
+  - Actual: The block title is used and the header title is ignored without any comparison or diagnostics.  
+  - Proposed: Compare the plaintext values, warn when redundant, and keep emitting hints when neither title form is present.

build.zig

@@ -10,6 +10,21 @@ const snapshot_files: []const []const u8 = &.{
     "test/snapshot/footnotes.hdoc",
 };
 
+const conformance_accept_files: []const []const u8 = &.{
+    "test/conformance/accept/inline_escape.hdoc",
+    "test/conformance/accept/title_header_redundant.hdoc",
+};
+
+const conformance_reject_files: []const []const u8 = &.{
+    "test/conformance/reject/string_cr_escape.hdoc",
+    "test/conformance/reject/inline_identifier_dash.hdoc",
+    "test/conformance/reject/heading_sequence.hdoc",
+    "test/conformance/reject/nested_top_level.hdoc",
+    "test/conformance/reject/container_children.hdoc",
+    "test/conformance/reject/time_relative_fmt.hdoc",
+    "test/conformance/reject/ref_in_heading.hdoc",
+};
+
 pub fn build(b: *std.Build) void {
     // Options:
     const target = b.standardTargetOptions(.{});
@@ -74,6 +89,40 @@ pub fn build(b: *std.Build) void {
         }
     }
 
+    // Conformance snapshots: accept cases (YAML only):
+    for (conformance_accept_files) |path| {
+        std.debug.assert(std.mem.endsWith(u8, path, ".hdoc"));
+        const yaml_file = b.fmt("{s}.yaml", .{path[0 .. path.len - 5]});
+
+        const test_run = b.addRunArtifact(exe);
+        test_run.addArgs(&.{ "--format", "yaml" });
+        test_run.addFileArg(b.path(path));
+        const generated_file = test_run.captureStdOut();
+
+        const compare_run = b.addRunArtifact(snapshot_diff);
+        compare_run.addFileArg(b.path(yaml_file));
+        compare_run.addFileArg(generated_file);
+
+        test_step.dependOn(&compare_run.step);
+    }
+
+    // Conformance snapshots: reject cases (diagnostics on stderr, expect exit code 1):
+    for (conformance_reject_files) |path| {
+        std.debug.assert(std.mem.endsWith(u8, path, ".hdoc"));
+        const diag_file = b.fmt("{s}.diag", .{path[0 .. path.len - 5]});
+
+        const test_run = b.addRunArtifact(exe);
+        test_run.addFileArg(b.path(path));
+        test_run.expectExitCode(1);
+        const generated_diag = test_run.captureStdErr();
+
+        const compare_run = b.addRunArtifact(snapshot_diff);
+        compare_run.addFileArg(b.path(diag_file));
+        compare_run.addFileArg(generated_diag);
+
+        test_step.dependOn(&compare_run.step);
+    }
+
     // Unit tests:
     const exe_tests = b.addTest(.{
         .root_module = b.createModule(.{

docs/TODO.md

@@ -5,30 +5,11 @@
 - Assign semantics to node types, paragraph kinds, ...
 - Specify "syntax" proper
 - Add links to RFCs where possible
-- Verbatim-body to text conversion is under-specified. You define verbatim syntax (: with | lines) and later say verbatim bodies become inline text spans (§8.2), but you don’t precisely define how piped lines join (LF vs preserving original CRLF, whether there is a trailing newline, whether a final EOF line_terminator contributes a newline, etc.). Different implementations may diverge. 
-- Inline “groups” exist syntactically but are not given explicit semantics. The grammar includes inline_group ::= "{" , inline_content , "}" and §5.4 makes brace balancing a core rule, but §8.2 doesn’t explicitly state that groups are semantically transparent (flattened) versus affecting whitespace normalization boundaries or span merging. 
 - Span attribute semantics are referenced but not fully defined. §8.2 introduces spans with an “attribute set (e.g. emphasis/monospace/link…)” but the spec never fully defines the canonical attribute keys, nesting behavior (e.g., \em inside \mono), or how lang overrides interact at span level. That’s a major interoperability risk because renderers may differ even if parsers agree. 
 - Refine that `hdoc(title)` is metadata while `title{}` is rendered rich text
-- Refine `img(path)` only using forward slash.
-  - Proposal: Add to §9.3.5:
-    - "path MUST use forward slashes (/) as path separators, regardless of host OS."
-    - "path MUST be relative; absolute paths and URI schemes (e.g., http://) MUST be rejected."
-    - "Path resolution is relative to the directory containing the HyperDoc source file."
-    - "Path traversal outside the source directory (e.g., ../../etc/passwd) SHOULD be rejected or restricted by implementations."
-- Proposal: Add to §9.2.4:
-  - "Multiple toc elements MAY appear in a document; each MUST render the same heading structure but MAY appear at different locations."
-  - "If depth differs between instances, each TOC renders independently according to its own depth attribute."
-- Add to §9.2.5:
-  - "Multiple footnotes elements partition footnote rendering; each instance collects only footnotes/citations accumulated since the previous dump (or document start)."
-- Proposal: Add to §4:
-  - "Implementations MUST support nesting depths of at least 32 levels."
-  - "Implementations MAY reject documents exceeding this depth with a diagnostic."
-  - "Nesting depth is measured as the maximum distance from the document root to any leaf node."
 - Ambiguity of Inline Unicode:
   - Finding: String literals ("...") support \u{...} escapes (§7.2.1). Inline text streams (bodies of p, h1) do not (§6.1 only lists \\, \{, \}).
   - Issue: Authors cannot enter invisible characters (like Non-Breaking Space U+00A0 or Zero Width Space U+200B) into a paragraph without pasting the raw invisible character, which is brittle and invisible in editors.
-- Recommendation: Add explicit sequencing in §7 stating: "Escape decoding MUST occur during semantic validation, before inline text construction (§8.2) for inline-list bodies, and before attribute validation for attribute values."
-- Recommendation: Add to §9.2.1: "If the document contains any \date, \time, or \datetime elements with fmt values other than iso, and hdoc(lang) is not specified, implementations SHOULD emit a diagnostic."
 - Issue: "Lexical" implies only regex-level matching. It does not strictly forbid 2023-02-31. For a strict format, "Semantic" validity (Gregorian correctness) should be enforced to prevent invalid metadata.
 
 ## Potential Future Features
@@ -120,4 +101,4 @@ quote {
 - `include(path="...")` is rejected for unbounded document content growth
 - `code` is just `\mono(syntax="…")`
 - `details/summary` is just HTML with dynamic changing page layout, ever tried printing this?
-- `\math`, `equation{…}` have too high implementation complexity and have high requirements on fonts, font renderers and layout engines.
+- `\math`, `equation{…}` have too high implementation complexity and have high requirements on fonts, font renderers and layout engines.

docs/specification.md

@@ -206,6 +206,12 @@ The grammar is intentionally ambiguous; a deterministic external rule selects a
 - Attribute values are **string literals** (see §5.5).
 - Attribute keys are identifiers with hyphen-separated segments (see §5.1 and §10.1).
 
+### 4.4 Nesting depth (syntax)
+
+- Implementations **MUST** support nesting depths of at least 32 levels.
+- Implementations **MAY** reject documents that exceed this depth with a diagnostic.
+- Nesting depth is measured as the maximum distance from the document root to any leaf node.
+
 ## 5. Grammar and additional syntax rules
 
 ### 5.1 Grammar (EBNF)
@@ -351,6 +357,8 @@ Tooling that aims to preserve author intent **SHOULD** preserve whether braces w
 
 Escape sequences are recognized only in string literals (node bodies of the `"..."` form and attribute values). No other syntax performs string-literal escape decoding.
 
+Escape decoding **MUST** occur during semantic validation, before inline text construction (§8.2) for inline-list bodies, and before attribute validation for attribute values.
+
 ### 7.1 Control character policy (semantic)
 
 - A semantic validator **MAY** reject TAB (U+0009) in source text.
@@ -432,6 +440,8 @@ Semantic processing **MUST** construct inline text as a sequence of **spans**, w
 - a Unicode string, and
 - an attribute set (e.g. emphasis/monospace/link, language overrides, etc.).
 
+Inline groups are structural only: when converting the inline tree into spans, implementations **MUST** flatten `inline_group` boundaries. An `inline_group` **MUST NOT** create a span boundary and **MUST NOT** affect whitespace normalization, but it **MUST** contribute the literal `{` and `}` characters to the inline text at its start and end.
+
 Processing rules:
 
 1. **Parse → tree:** Parsing preserves `ws` and yields an inline tree (text items, inline nodes, and inline groups).
@@ -586,6 +596,9 @@ The elements in this chapter **MUST** appear only as top-level block elements (d
   - `date` (optional): datetime lexical format (§10.2.3)
   - `tz` (optional): default timezone for time/datetime values (§10.2)
 
+Diagnostics:
+- If the document contains any `\date`, `\time`, or `\datetime` elements with `fmt` values other than `iso`, and `hdoc(lang)` is not specified, implementations **SHOULD** emit a diagnostic.
+
 #### 9.2.2 `title` (document title)
 
 - **Role:** document-level display title
@@ -624,6 +637,8 @@ Heading structure and numbering:
 
 Semantic constraints:
 - `toc` **MUST** be a top-level block element (a direct child of the document).
+- Multiple `toc` elements **MAY** appear in a document; each **MUST** render the same heading structure but **MAY** appear at different locations.
+- If `depth` differs between instances, each `toc` **MUST** render independently according to its own `depth` attribute.
 
 #### 9.2.5 Footnote dump: `footnotes`
 
@@ -635,6 +650,7 @@ Semantic constraints:
 
 Semantics:
 
+- Multiple `footnotes` elements **MAY** appear in a document.
 - `footnotes;` collects and renders all footnotes of all kinds accumulated since the previous `footnotes(...)` node (or since start of document if none appeared yet).
 - `footnotes(kind="footnote");` collects and renders only `kind="footnote"` entries accumulated since the previous `footnotes(...)` node.
 - `footnotes(kind="citation");` collects and renders only `kind="citation"` entries accumulated since the previous `footnotes(...)` node.
@@ -686,6 +702,13 @@ Only an empty body (`;`) is not "inline text".
   - `lang` (optional)
   - `id` (optional; top-level only)
 
+Path semantics:
+
+- `path` **MUST** use forward slashes (`/`) as path separators, regardless of host operating system.
+- `path` **MUST** be relative; absolute paths and URI schemes **MUST** be rejected.
+- Path resolution is relative to the directory containing the HyperDoc source file.
+- Path traversal outside the source directory (e.g., `../../etc/passwd`) **SHOULD** be rejected or restricted by implementations.
+
 #### 9.3.6 Preformatted: `pre`
 
 - **Body:** inline text