Support Pandoc escaped spaces and end-of-block backslash hard breaks

README.md

@@ -15,6 +15,7 @@ xhtmlmd is largely implemented using AI, except for the tests. The tests are lar
 - HTML-in-Markdown: block containers opened with `markdown="1"`; the control attribute is stripped, indented code blocks are disabled inside the container, and fenced code is the code-block syntax there.
 - Math: four modes: `brackets` for `\(...\)`, `\[...\]`, and `$$...$$`, `dollars` for those plus `$...$` using Pandoc's non-space/digit dollar rules, `on` to preserve `\(...\)` and `\[...\]` delimiters for client-side renderers such as KaTeX, and `off`. Brackets mode is the default.
 - Attributes and inline spans: Pandoc/kramdown-style `{#id .class key="value"}`, block IALs `{: ...}`, span IALs, ALDs such as `{:note: #id .class}` with references, superscript `^x^`, subscript `~x~`, and highlight `==x==`.
+- Pandoc-style escapes: `\ ` is a non-breaking space, and a trailing `\` is a hard break even at the end of a block, so a paragraph of just `\` renders as a `<br />` spacer.
 - Definition lists: PHP Markdown Extra/Pandoc-style `Term` followed by `: definition` or `~ definition`.
 - Footnotes: `[^id]` references to defined `[^id]:` definitions with indented continuation blocks.
 - Abbreviations: `*[HTML]: Hyper Text Markup Language` definitions render matching text as `<abbr>`.

docs/DIALECT.md

@@ -8,6 +8,7 @@ When Markdown extensions disagree, this crate chooses the behavior closest to Pa
 - Definition lists follow PHP Markdown Extra/Pandoc: one-line terms with one or more `:` or `~` definitions.
 - Footnotes follow Pandoc/kramdown label rules and render as XHTML endnotes with backlinks. The endnotes `<section>` has no leading `<hr>` (unlike cmark-gfm): separators are a styling concern, so add one with CSS if wanted.
 - Inline `~~x~~` renders as strikethrough. Inline `~x~` renders as subscript, using the same no-whitespace rule as superscript `^x^`.
+- Backslash escapes follow Pandoc's `escaped_line_breaks`/escaped-space rules: `\ ` (backslash before a space) is a non-breaking space, and a backslash at the end of a line is a hard break even at the end of a block, so a paragraph of just `\` renders as `<p><br /></p>`. CommonMark instead keeps both as literal text. An escaped backslash (`\\`) never starts a break.
 - `<tag markdown="1">` parses block Markdown inside the balanced tag. `markdown="span"` parses inline content into a single paragraph child.
 - Raw HTML passes through unbalanced, per CommonMark. The opt-in `balance` option closes unclosed elements at the fragment end, drops stray closes, and self-closes void tags, without HTML5 implied-end-tag rules.
 - Math defaults to `MathMode::Brackets`, which recognizes `\(...\)`, `\[...\]`, and `$$...$$`. `MathMode::Dollars` also recognizes `$...$` with Pandoc's guard against currency-like spans. `MathMode::On` preserves backslashes before `[]()` so client-side renderers such as KaTeX can see TeX delimiters. `MathMode::Off` treats TeX delimiters as ordinary Markdown text.

src/inline.rs

@@ -42,6 +42,7 @@ fn parse_inner(src: &str, ctx: &InlineContext<'_>, depth: usize) -> Vec<Inline>
     };
     let mut failed = FailedScans::default();
     let mut i = 0;
+    let mut last_escape_end = usize::MAX;
     while i < src.len() {
         if starts(src, i, "\\[")
             && matches!(ctx.options.math, MathMode::Brackets | MathMode::Dollars)
@@ -210,24 +211,33 @@ fn parse_inner(src: &str, ctx: &InlineContext<'_>, depth: usize) -> Vec<Inline>
         if starts(src, i, "\\") {
             if i + 1 < src.len() {
                 let next = next_char(src, i + 1);
+                if next == ' ' {
+                    scanner.text.push('\u{a0}');
+                    i += 2;
+                    last_escape_end = i;
+                    continue;
+                }
                 if is_escapable(next) {
                     if ctx.options.math == MathMode::On && matches!(next, '[' | ']' | '(' | ')') {
                         scanner.text.push('\\');
                         scanner.text.push(next);
                         i += 1 + next.len_utf8();
+                        last_escape_end = i;
                         continue;
                     }
                     scanner
                         .text
                         .push(if next == '&' { ESCAPED_AMP } else { next });
                     i += 1 + next.len_utf8();
+                    last_escape_end = i;
                     continue;
                 }
             }
         }
         if ch == '\n' {
-            if scanner.text.ends_with("  ") || scanner.text.ends_with('\\') {
-                if scanner.text.ends_with('\\') {
+            let backslash_break = scanner.text.ends_with('\\') && last_escape_end != i;
+            if backslash_break || scanner.text.ends_with("  ") {
+                if backslash_break {
                     scanner.text.pop();
                 } else {
                     while scanner.text.ends_with(' ') {
@@ -246,6 +256,11 @@ fn parse_inner(src: &str, ctx: &InlineContext<'_>, depth: usize) -> Vec<Inline>
             i += ch.len_utf8();
         }
     }
+    if scanner.text.ends_with('\\') && last_escape_end != src.len() {
+        scanner.text.pop();
+        scanner.flush_text();
+        scanner.push_inline(Inline::HardBreak);
+    }
     scanner.flush_text();
     process_delimiters(&mut nodes, &mut delimiters);
     nodes_to_inlines(&nodes)

tests/source/cmark-gfm/spec.txt

@@ -1155,7 +1155,7 @@ Foo
 
 Nor does a backslash at the end:
 
-```````````````````````````````` example
+```````````````````````````````` example disabled
 Foo\
 ----
 .
@@ -5786,7 +5786,7 @@ Any ASCII punctuation character may be backslash-escaped:
 Backslashes before other characters are treated as literal
 backslashes:
 
-```````````````````````````````` example
+```````````````````````````````` example disabled
 \→\A\a\ \3\φ\«
 .
 <p>\→\A\a\ \3\φ\«</p>
@@ -9770,7 +9770,7 @@ Hard line breaks are for separating inline content within a block.
 Neither syntax for hard line breaks works at the end of a paragraph or
 other block element:
 
-```````````````````````````````` example
+```````````````````````````````` example disabled
 foo\
 .
 <p>foo\</p>
@@ -9784,7 +9784,7 @@ foo
 ````````````````````````````````
 
 
-```````````````````````````````` example
+```````````````````````````````` example disabled
 ### foo\
 .
 <h3>foo\</h3>

tests/test_focused.py

@@ -90,6 +90,21 @@ def test_balance_ignores_rawtext_and_voids():
     assert "</div>" not in html
     assert "<br />" in html
 
+def test_escaped_space_is_nbsp():
+    assert to_xhtml("a\\ b\n") == "<p>a\u00a0b</p>\n"
+    assert to_xhtml("# foo\\ bar\n") == "<h1>foo\u00a0bar</h1>\n"
+
+def test_trailing_backslash_is_hard_break_at_block_end():
+    assert to_xhtml("text\\\n\nnext\n") == "<p>text<br />\n</p>\n<p>next</p>\n"
+    assert to_xhtml("\\\n") == "<p><br />\n</p>\n"
+    assert to_xhtml("\\ \n") == "<p><br />\n</p>\n"
+    assert to_xhtml("# foo\\\n") == "<h1>foo<br />\n</h1>\n"
+    assert to_xhtml("Foo\\\n----\n") == "<h2>Foo<br />\n</h2>\n"
+
+def test_escaped_backslash_is_never_a_break():
+    assert to_xhtml("a\\\\\nb\n") == "<p>a\\\nb</p>\n"
+    assert to_xhtml("a\\\\\n") == "<p>a\\</p>\n"
+
 def test_long_nonascii_words_near_autolink_cap_do_not_error():
     for boundary in ("(", "a: ", "x '"):
         for count in (126, 127, 128, 129, 130, 200):