feat(ci-new): add markdown structural comparison commands

Add `markdown-compare` and `markdown-compare-all` commands to verify
that translated docs maintain the same structure as the reference.
Compare heading levels, code fences, lists, quotes, and other
structural elements while allowing text differences.

7 files changed, 439 insertions, 11 deletions

diff --git a/mingling_ci/help.txt b/mingling_ci/help.txt
index 8dbc75b..a810b41 100644
--- a/mingling_ci/help.txt
+++ b/mingling_ci/help.txt
@@ -9,15 +9,17 @@ FLAGS:
 
 COMMANDS:
   UTILS:
-   markdown-check <PATH>    Verify rust code blocks in one markdown file
-   markdown-check-all       Verify rust code blocks in all configured markdown files
-   report-collect           Collect and organize all inspection reports
-   report-clean             Clean up all reports
+   report-collect                       Collect and organize all inspection reports
+   report-clean                         Clean up all reports
 
-   show-features            Print the docs.rs feature list of mingling
-   show-manifests           Print all crate paths that need to be checked
+   show-features                        Print the docs.rs feature list of mingling
+   show-manifests                       Print all crate paths that need to be checked
 
   TASKS:
-   build-all                Build all crates
-   clippy-all               Run clippy with -D warnings on all crates
-   test-all                 Test all crates
+   markdown-check                       Verify rust code blocks in one markdown file
+   markdown-check-all <PATH>            Verify rust code blocks in all configured markdown files
+   markdown-compare                     Compare the structure of two markdown files/dirs
+   markdown-compare-all <PATH> <PATH>   Compare all translated docs against the reference
+   build-all                            Build all crates
+   clippy-all                           Run clippy with -D warnings on all crates
+   test-all                             Test all crates
diff --git a/mingling_ci/src/cmd.rs b/mingling_ci/src/cmd.rs
index b531b92..30d65a1 100644
--- a/mingling_ci/src/cmd.rs
+++ b/mingling_ci/src/cmd.rs
@@ -1,4 +1,3 @@
-pub(crate) mod cmd_markdown_check;
 pub(crate) mod cmd_report_clean;
 pub(crate) mod cmd_report_collect;
 pub(crate) mod cmd_show_features;
diff --git a/mingling_ci/src/markdown.rs b/mingling_ci/src/markdown.rs
index b231a89..75f2cbe 100644
--- a/mingling_ci/src/markdown.rs
+++ b/mingling_ci/src/markdown.rs
@@ -1,2 +1,3 @@
+pub(crate) mod compare;
 pub(crate) mod project;
 pub(crate) mod test;
diff --git a/mingling_ci/src/markdown/compare.rs b/mingling_ci/src/markdown/compare.rs
new file mode 100644
index 0000000..1bf3c57
--- /dev/null
+++ b/mingling_ci/src/markdown/compare.rs
@@ -0,0 +1,203 @@
+//! Structural comparison of markdown docs (reference vs translation).
+//!
+//! For each file pair the comparison uses a *structural signature*: one token
+//! per line, classifying headings (both Markdown `#` and HTML `<hN>`), fenced
+//! code blocks (including their language tag), `@@@` hidden-compilation lines,
+//! blank lines, blockquotes, lists and plain text. Translated text is allowed
+//! to differ; the structure is not.
+
+use std::path::{Path, PathBuf};
+
+/// Collects all `.md` files under `dir`, returned relative to it.
+pub(crate) fn collect_md_files(dir: &Path) -> Vec<PathBuf> {
+    let mut out = Vec::new();
+    let mut stack = vec![dir.to_path_buf()];
+    while let Some(current) = stack.pop() {
+        let Ok(entries) = std::fs::read_dir(&current) else {
+            continue;
+        };
+        for entry in entries.flatten() {
+            let path = entry.path();
+            if path.is_dir() {
+                stack.push(path);
+            } else if path.extension().is_some_and(|e| e == "md") {
+                out.push(path.strip_prefix(dir).unwrap_or(&path).to_path_buf());
+            }
+        }
+    }
+    out.sort();
+    out
+}
+
+/// Compares the structural signatures of two markdown files.
+///
+/// Returns the human-readable diff lines (up to a small window) on the first
+/// structural difference.
+pub(crate) fn compare_signature(ref_path: &Path, lang_path: &Path) -> Result<(), Vec<String>> {
+    let ref_content = std::fs::read_to_string(ref_path).unwrap_or_default();
+    let lang_content = std::fs::read_to_string(lang_path).unwrap_or_default();
+
+    let ref_sig = signature_of(&ref_content);
+    let lang_sig = signature_of(&lang_content);
+
+    if ref_sig == lang_sig {
+        return Ok(());
+    }
+
+    let ref_lines: Vec<&str> = ref_content.lines().collect();
+    let lang_lines: Vec<&str> = lang_content.lines().collect();
+
+    let mut diffs = Vec::new();
+    let mut window = 0;
+    let max = ref_sig.len().max(lang_sig.len());
+    for i in 0..max {
+        let ref_tok = ref_sig.get(i);
+        let lang_tok = lang_sig.get(i);
+        if ref_tok == lang_tok {
+            continue;
+        }
+        if window >= 5 {
+            diffs.push(format!("... ({}-line window truncated)", max - i));
+            break;
+        }
+        window += 1;
+        let ref_line = ref_lines.get(i).copied().unwrap_or("<missing>");
+        let lang_line = lang_lines.get(i).copied().unwrap_or("<missing>");
+        diffs.push(format!("line {}", i + 1));
+        diffs.push(format!(
+            "expect `{}` {}",
+            token_label(ref_tok.map_or("<eof>", String::as_str)),
+            display_line(ref_line)
+        ));
+        diffs.push(format!(
+            "found  `{}` {}",
+            token_label(lang_tok.map_or("<eof>", String::as_str)),
+            display_line(lang_line)
+        ));
+        if ref_sig.len() != lang_sig.len() && window >= 5 {
+            diffs.push(format!(
+                "note: reference has {} lines, translation has {} lines",
+                ref_sig.len(),
+                lang_sig.len()
+            ));
+            break;
+        }
+    }
+    if diffs.is_empty() {
+        diffs.push("signatures differ in length (see line count note)".to_string());
+    }
+    Err(diffs)
+}
+
+/// Builds the structural signature of a markdown file.
+fn signature_of(content: &str) -> Vec<String> {
+    let mut sig = Vec::new();
+    let mut in_fence = false;
+    let mut fence_lang = String::new();
+
+    for raw_line in content.lines() {
+        let line = raw_line.trim();
+
+        if in_fence {
+            if line.starts_with("```") {
+                in_fence = false;
+                sig.push(format!("F:{fence_lang}"));
+            } else if line.starts_with("@@@") {
+                sig.push("A".to_string());
+            } else if line.is_empty() {
+                sig.push("B".to_string());
+            } else {
+                sig.push("P".to_string());
+            }
+            continue;
+        }
+
+        if line.starts_with("```") {
+            in_fence = true;
+            fence_lang = line.trim_start_matches("```").trim().to_string();
+            sig.push(format!("F:{fence_lang}"));
+        } else if line.starts_with('#') {
+            let level = line.chars().take_while(|c| *c == '#').count();
+            sig.push(format!("H{level}"));
+        } else if line.starts_with("<h") || line.starts_with("</h") {
+            // HTML headings (e.g. `<h1 align="center">` / `</h1>`)
+            let level = line
+                .trim_start_matches(['<', '/'])
+                .chars()
+                .next()
+                .and_then(|c| c.to_digit(10))
+                .unwrap_or(1);
+            sig.push(format!("H{level}"));
+        } else if line.starts_with("@@@") {
+            sig.push("A".to_string());
+        } else if line.is_empty() {
+            sig.push("B".to_string());
+        } else if line.starts_with('>') {
+            sig.push("Q".to_string());
+        } else if is_list_line(line) {
+            sig.push("L".to_string());
+        } else {
+            sig.push("P".to_string());
+        }
+    }
+    sig
+}
+
+/// Human-readable label for a structural token.
+fn token_label(token: &str) -> String {
+    match token {
+        "B" => "blank".to_string(),
+        "A" => "@@@".to_string(),
+        "Q" => "quote".to_string(),
+        "L" => "list".to_string(),
+        "P" => "text".to_string(),
+        t if t.starts_with('H') => format!("heading-{}", &t[1..]),
+        t if t.starts_with("F:") => {
+            let lang = &t[2..];
+            if lang.is_empty() {
+                "fence".to_string()
+            } else {
+                format!("fence:{lang}")
+            }
+        }
+        _ => token.to_string(),
+    }
+}
+
+/// Renders a source line for display: blank lines become `<blank>`.
+fn display_line(line: &str) -> String {
+    if line.trim().is_empty() {
+        "<blank>".to_string()
+    } else {
+        truncate(line)
+    }
+}
+
+fn truncate(line: &str) -> String {
+    const MAX: usize = 60;
+    if line.chars().count() <= MAX {
+        line.to_string()
+    } else {
+        let cut: String = line.chars().take(MAX).collect();
+        format!("{cut}...")
+    }
+}
+
+fn is_list_line(line: &str) -> bool {
+    let trimmed = line.trim_start();
+    trimmed.starts_with("- ")
+        || trimmed.starts_with("* ")
+        || trimmed.starts_with("+ ")
+        || is_numbered_list(trimmed)
+}
+
+/// A numbered list item: `1. text`, `1) text`, `10. text`, ...
+fn is_numbered_list(line: &str) -> bool {
+    let digit_count = line.chars().take_while(char::is_ascii_digit).count();
+    if digit_count == 0 {
+        return false;
+    }
+    let rest = &line[digit_count..];
+    (rest.starts_with(". ") || rest.starts_with(") "))
+        && rest.chars().nth(1).is_some_and(|c| c == ' ' || c == '\t')
+}
diff --git a/mingling_ci/src/task.rs b/mingling_ci/src/task.rs
index b5161ad..92660bb 100644
--- a/mingling_ci/src/task.rs
+++ b/mingling_ci/src/task.rs
@@ -1,4 +1,6 @@
 pub(crate) mod cmd_build;
 pub(crate) mod cmd_clippy;
+pub(crate) mod cmd_markdown_check;
+pub(crate) mod cmd_markdown_compare;
 pub(crate) mod cmd_test;
 pub(crate) mod run;
diff --git a/mingling_ci/src/cmd/cmd_markdown_check.rs b/mingling_ci/src/task/cmd_markdown_check.rs
index 6044846..2408636 100644
--- a/mingling_ci/src/cmd/cmd_markdown_check.rs
+++ b/mingling_ci/src/task/cmd_markdown_check.rs
@@ -74,7 +74,7 @@ pub async fn markdown_check_all() -> Next {
 }
 
 /// The file name without extension, e.g. `README.md` → `README`.
-fn stem_of(path: &Path) -> String {
+pub(crate) fn stem_of(path: &Path) -> String {
     path.file_stem()
         .unwrap_or_default()
         .to_string_lossy()
diff --git a/mingling_ci/src/task/cmd_markdown_compare.rs b/mingling_ci/src/task/cmd_markdown_compare.rs
new file mode 100644
index 0000000..b014f1e
--- /dev/null
+++ b/mingling_ci/src/task/cmd_markdown_compare.rs
@@ -0,0 +1,221 @@
+use std::collections::BTreeSet;
+use std::path::{Path, PathBuf};
+
+use colored::Colorize;
+use just_fmt::snake_case;
+use mingling::{
+    Grouped, Routable,
+    macros::{buffer, command, renderer},
+    res::ResExitCode,
+};
+
+use crate::Next;
+use crate::markdown::compare::{collect_md_files, compare_signature};
+use crate::reporter::{self, ReportResult};
+use crate::task::cmd_markdown_check::{ErrorMarkdownArgs, ErrorMarkdownConfig, stem_of};
+
+const DOCS_DIR: &str = "./docs";
+const LANG_CONFIG: &str = ".config/docs-lang.txt";
+
+/// One file-pair outcome of a structure comparison.
+struct CompareOutcome {
+    item: String,
+    location: String,
+    ok: bool,
+    output: String,
+}
+
+#[command(node = "markdown-compare")]
+// `#[command]` rewrites an owned first param into the entry type, so the args
+// must be passed by value even though the body only reads them.
+#[allow(clippy::needless_pass_by_value)]
+pub fn markdown_compare(args: Vec<String>) -> Next {
+    let [ref_arg, trans_arg] = args.as_slice() else {
+        return ErrorMarkdownArgs("missing <reference> and <translation> arguments".to_string())
+            .to_chain();
+    };
+    let ref_path = cwd().join(ref_arg);
+    let trans_path = cwd().join(trans_arg);
+
+    reporter::set_task("Markdown-Compare");
+    let outcomes = if ref_path.is_dir() && trans_path.is_dir() {
+        compare_dirs(&ref_path, &trans_path, "doc")
+    } else if ref_path.is_file() && trans_path.is_file() {
+        compare_files(&ref_path, &trans_path, "doc")
+    } else {
+        return ErrorMarkdownArgs(
+            "both arguments must be files or both must be directories".to_string(),
+        )
+        .to_chain();
+    };
+    let fail_count = export_outcomes(&outcomes);
+    reporter::flush();
+
+    ResultMarkdownCompare { fail_count }.to_chain()
+}
+
+#[command(node = "markdown-compare-all")]
+pub fn markdown_compare_all() -> Next {
+    let Some(langs) = lang_config() else {
+        return ErrorMarkdownConfig.to_chain();
+    };
+    let Some(reference) = langs.first() else {
+        return ErrorMarkdownConfig.to_chain();
+    };
+    let ref_dir = PathBuf::from(DOCS_DIR).join(reference);
+    if !ref_dir.is_dir() {
+        return ErrorMarkdownArgs(format!(
+            "reference docs directory `{}` does not exist",
+            ref_dir.display()
+        ))
+        .to_chain();
+    }
+
+    reporter::set_task("Markdown-Compare-All");
+    let mut fail_count = 0;
+    for lang in &langs[1..] {
+        let lang_dir = PathBuf::from(DOCS_DIR).join(lang);
+        if !lang_dir.is_dir() {
+            eprintln!(
+                "  {}: `{}` does not exist",
+                "ERROR".bright_red(),
+                lang_dir.display()
+            );
+            fail_count += 1;
+            continue;
+        }
+        let outcomes = compare_dirs(&ref_dir, &lang_dir, &lang_key(lang));
+        fail_count += export_outcomes(&outcomes);
+    }
+    reporter::flush();
+
+    ResultMarkdownCompare { fail_count }.to_chain()
+}
+
+/// Compares one file pair (reference vs translation).
+fn compare_files(ref_path: &Path, trans_path: &Path, prefix: &str) -> Vec<CompareOutcome> {
+    let item = format!("{prefix}-{}", snake_case!(&stem_of(ref_path)));
+    let location = trans_path.to_string_lossy().into_owned();
+    match compare_signature(ref_path, trans_path) {
+        Ok(()) => vec![CompareOutcome {
+            item,
+            location,
+            ok: true,
+            output: String::new(),
+        }],
+        Err(diffs) => vec![CompareOutcome {
+            item,
+            location,
+            ok: false,
+            output: diffs.join("\n"),
+        }],
+    }
+}
+
+/// Compares two directories: every `.md` file in the reference must exist in
+/// the translation with the same structural signature; extra files are errors.
+fn compare_dirs(ref_dir: &Path, trans_dir: &Path, prefix: &str) -> Vec<CompareOutcome> {
+    let ref_files = collect_md_files(ref_dir);
+    let ref_set: BTreeSet<PathBuf> = ref_files.iter().cloned().collect();
+    let trans_set: BTreeSet<PathBuf> = collect_md_files(trans_dir).into_iter().collect();
+
+    let mut outcomes = Vec::new();
+    for file in ref_files {
+        let item = format!("{prefix}-{}", snake_case!(&stem_of(&file)));
+        let trans_path = trans_dir.join(&file);
+        let location = trans_path.to_string_lossy().into_owned();
+        if !trans_set.contains(&file) {
+            outcomes.push(CompareOutcome {
+                item,
+                location,
+                ok: false,
+                output: "missing in translation".to_string(),
+            });
+            continue;
+        }
+        outcomes.push(match compare_signature(&ref_dir.join(&file), &trans_path) {
+            Ok(()) => CompareOutcome {
+                item,
+                location,
+                ok: true,
+                output: String::new(),
+            },
+            Err(diffs) => CompareOutcome {
+                item,
+                location,
+                ok: false,
+                output: diffs.join("\n"),
+            },
+        });
+    }
+
+    for file in trans_set.difference(&ref_set) {
+        let item = format!("{prefix}-{}", snake_case!(&stem_of(file)));
+        let trans_path = trans_dir.join(file);
+        outcomes.push(CompareOutcome {
+            item,
+            location: trans_path.to_string_lossy().into_owned(),
+            ok: false,
+            output: "extra file, not in reference".to_string(),
+        });
+    }
+    outcomes
+}
+
+/// Exports the outcomes via `reporter`; failures also print to stderr.
+fn export_outcomes(outcomes: &[CompareOutcome]) -> usize {
+    let mut fail_count = 0;
+    for outcome in outcomes {
+        if outcome.ok {
+            reporter::export(&outcome.item, &outcome.location, ReportResult::Ok);
+        } else {
+            fail_count += 1;
+            eprintln!("  {} {}", "failed".bright_red(), outcome.item);
+            eprintln!("  {}\n{}", outcome.location, outcome.output);
+            reporter::export(
+                &outcome.item,
+                &outcome.location,
+                ReportResult::Error(outcome.output.clone()),
+            );
+        }
+    }
+    fail_count
+}
+
+/// Reads `.config/docs-lang.txt`: the first line is the reference directory
+/// (relative to `./docs/`), the rest are translations that must mirror it.
+fn lang_config() -> Option<Vec<String>> {
+    let content = std::fs::read_to_string(LANG_CONFIG).ok()?;
+    Some(
+        content
+            .lines()
+            .map(str::trim)
+            .filter(|l| !l.is_empty() && !l.starts_with('#'))
+            .map(|l| l.trim_start_matches("./").to_string())
+            .collect(),
+    )
+}
+
+/// Turns a lang directory path into a report-item key, e.g.
+/// `./_zh_CN/pages/` → `_zh_CN_pages`.
+fn lang_key(lang: &str) -> String {
+    lang.trim_matches('/').replace('/', "_")
+}
+
+fn cwd() -> PathBuf {
+    std::env::current_dir().unwrap_or_else(|_| PathBuf::from("."))
+}
+
+/// Number of files that failed the structure comparison.
+#[derive(Grouped)]
+pub struct ResultMarkdownCompare {
+    pub fail_count: usize,
+}
+
+/// Silently sets a non-zero exit code when any comparison failed.
+#[renderer(buffer)]
+pub fn render_markdown_compare(r: ResultMarkdownCompare, exit_code: &mut ResExitCode) {
+    if r.fail_count > 0 {
+        exit_code.exit_code = 1;
+    }
+}