ANW-18 doctor base checks + FrontmatterNotMapping variant

src/doctor.rs

@@ -0,0 +1,195 @@
+//! `anwesen doctor` base checks per [ANW-18]. Walks the vault once,
+//! collects hard failures and soft warnings from [`crate::vault::scan`],
+//! plus any HTTP-surface path collisions that scrape past the OS path
+//! distinction (e.g. Unicode normalization differences). Returns a
+//! [`Report`] the binary renders to stdout; a non-empty report exits
+//! non-zero per the User Manual.
+
+use std::collections::BTreeMap;
+use std::fmt::Write;
+use std::path::Path;
+
+use crate::vault::{self, ScanIssue, ScanWarning};
+
+#[derive(Debug, Default)]
+pub struct Report {
+    pub note_count: usize,
+    pub issues: Vec<ScanIssue>,
+    pub warnings: Vec<ScanWarning>,
+    pub path_collisions: Vec<PathCollision>,
+}
+
+#[derive(Debug)]
+pub struct PathCollision {
+    pub path: String,
+    pub count: usize,
+}
+
+impl Report {
+    /// True if any anomaly was found -- the contract for the doctor exit
+    /// code per the User Manual.
+    #[must_use]
+    pub fn is_clean(&self) -> bool {
+        self.issues.is_empty() && self.warnings.is_empty() && self.path_collisions.is_empty()
+    }
+}
+
+/// Run `doctor`'s base checks on the given vault root. Pure -- the binary
+/// is responsible for rendering the report and choosing an exit code.
+#[must_use]
+pub fn run(vault_root: &Path) -> Report {
+    let scan = vault::scan(vault_root);
+    let collisions = detect_path_collisions(
+        &scan
+            .notes
+            .iter()
+            .map(|n| n.path.clone())
+            .collect::<Vec<_>>(),
+    );
+    Report {
+        note_count: scan.notes.len(),
+        issues: scan.issues,
+        warnings: scan.warnings,
+        path_collisions: collisions,
+    }
+}
+
+/// Find HTTP-surface path collisions: two distinct OS files whose
+/// vault-relative, forward-slash-normalized paths are equal. Almost
+/// never bites on Linux (the OS already enforces unique paths), but
+/// Unicode-NFC vs -NFD or symlink-stitched paths can in principle hit it.
+#[must_use]
+pub fn detect_path_collisions(paths: &[String]) -> Vec<PathCollision> {
+    let mut counts: BTreeMap<&str, usize> = BTreeMap::new();
+    for p in paths {
+        *counts.entry(p.as_str()).or_insert(0) += 1;
+    }
+    counts
+        .into_iter()
+        .filter(|(_, c)| *c > 1)
+        .map(|(p, c)| PathCollision {
+            path: p.to_string(),
+            count: c,
+        })
+        .collect()
+}
+
+/// Render the report to a multi-line string the binary writes to stdout.
+#[must_use]
+pub fn render(vault_root: &Path, report: &Report) -> String {
+    let mut out = String::new();
+    let _ = writeln!(out, "anwesen doctor: {}", vault_root.display());
+    let _ = writeln!(out, "  notes:           {}", report.note_count);
+    let _ = writeln!(out, "  issues:          {}", report.issues.len());
+    let _ = writeln!(out, "  warnings:        {}", report.warnings.len());
+    let _ = writeln!(out, "  path collisions: {}", report.path_collisions.len());
+    if !report.issues.is_empty() {
+        out.push_str("\nissues:\n");
+        for issue in &report.issues {
+            let _ = writeln!(out, "  {}: {}", issue.path.display(), issue.kind);
+        }
+    }
+    if !report.warnings.is_empty() {
+        out.push_str("\nwarnings:\n");
+        for w in &report.warnings {
+            let _ = writeln!(out, "  {}: {}", w.path.display(), w.kind);
+        }
+    }
+    if !report.path_collisions.is_empty() {
+        out.push_str("\npath collisions:\n");
+        for c in &report.path_collisions {
+            let _ = writeln!(out, "  {} ({} files)", c.path, c.count);
+        }
+    }
+    if report.is_clean() {
+        out.push_str("\nOK.\n");
+    } else {
+        out.push_str("\nFAIL.\n");
+    }
+    out
+}
+
+/// Convenience: run + render against an absolute vault path. Returns the
+/// rendered report and the exit code the binary should propagate.
+#[must_use]
+pub fn run_and_render(vault_root: &Path) -> (String, i32) {
+    let report = run(vault_root);
+    let exit = i32::from(!report.is_clean());
+    (render(vault_root, &report), exit)
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+    use std::fs;
+    use tempfile::TempDir;
+
+    fn write(root: &Path, rel: &str, body: &str) {
+        let p = root.join(rel);
+        if let Some(parent) = p.parent() {
+            fs::create_dir_all(parent).unwrap();
+        }
+        fs::write(p, body).unwrap();
+    }
+
+    #[test]
+    fn clean_vault_reports_clean() {
+        let tmp = TempDir::new().unwrap();
+        write(tmp.path(), "a.md", "---\ntags: [demo]\n---\nbody\n");
+        let r = run(tmp.path());
+        assert!(r.is_clean(), "expected clean: {r:?}");
+        assert_eq!(r.note_count, 1);
+    }
+
+    #[test]
+    fn malformed_frontmatter_reports_issue_not_warning() {
+        let tmp = TempDir::new().unwrap();
+        write(tmp.path(), "bad.md", "---\nkey: : :\n---\n");
+        let r = run(tmp.path());
+        assert!(!r.is_clean());
+        assert_eq!(r.note_count, 0);
+        assert_eq!(r.issues.len(), 1);
+        assert!(r.warnings.is_empty());
+    }
+
+    #[test]
+    fn frontmatter_not_mapping_reports_warning_keeps_note() {
+        let tmp = TempDir::new().unwrap();
+        write(tmp.path(), "list.md", "---\n- a\n- b\n---\nbody\n");
+        let r = run(tmp.path());
+        // Note is kept (serve compatibility); doctor flags the warning.
+        assert_eq!(r.note_count, 1);
+        assert!(r.issues.is_empty());
+        assert_eq!(r.warnings.len(), 1);
+        assert!(!r.is_clean());
+    }
+
+    #[test]
+    fn detect_path_collisions_finds_duplicates() {
+        let dups =
+            detect_path_collisions(&["a.md".into(), "b.md".into(), "a.md".into(), "a.md".into()]);
+        assert_eq!(dups.len(), 1);
+        assert_eq!(dups[0].path, "a.md");
+        assert_eq!(dups[0].count, 3);
+    }
+
+    #[test]
+    fn run_and_render_exits_zero_on_clean() {
+        let tmp = TempDir::new().unwrap();
+        write(tmp.path(), "a.md", "---\n---\n");
+        let (out, exit) = run_and_render(tmp.path());
+        assert_eq!(exit, 0);
+        assert!(out.contains("OK."));
+        assert!(out.contains("notes:           1"));
+    }
+
+    #[test]
+    fn run_and_render_exits_nonzero_on_issue() {
+        let tmp = TempDir::new().unwrap();
+        write(tmp.path(), "bad.md", "---\n:: :: ::\n---\n");
+        let (out, exit) = run_and_render(tmp.path());
+        assert_eq!(exit, 1);
+        assert!(out.contains("FAIL."));
+        assert!(out.contains("issues:"));
+    }
+}

src/lib.rs

@@ -5,6 +5,7 @@
 //! supervisor tree, and the HTTP surface as those issues land.
 
 pub mod app;
+pub mod doctor;
 pub mod health;
 pub mod http;
 pub mod index;

src/main.rs

@@ -6,6 +6,7 @@
 mod cli;
 
 use anwesen::app::Anwesen;
+use anwesen::doctor;
 use anyhow::Result;
 use clap::Parser;
 use hydra::Application;
@@ -31,10 +32,11 @@ fn main() -> Result<()> {
         }
         Command::Doctor(args) => {
             init_logging(args.log_level);
-            tracing::info!(
-                vault = %args.vault.display(),
-                "anwesen doctor: not yet implemented (ANW-10 stub)"
-            );
+            let (rendered, exit) = doctor::run_and_render(&args.vault);
+            // Render to stdout so the report is pipe-friendly; logs go to
+            // stderr via the tracing subscriber.
+            print!("{rendered}");
+            std::process::exit(exit);
         }
         Command::Version => {
             println!("{}", env!("CARGO_PKG_VERSION"));

src/vault.rs

@@ -112,7 +112,11 @@ pub fn frontmatter_to_json(fm: &Frontmatter) -> JsonValue {
 #[derive(Debug)]
 pub struct ScanResult {
     pub notes: Vec<Note>,
+    /// Hard failures -- file could not be loaded at all.
     pub issues: Vec<ScanIssue>,
+    /// Soft anomalies -- file loaded but flagged for `doctor`. `serve`
+    /// ignores these; ANW-18 surfaces them.
+    pub warnings: Vec<ScanWarning>,
 }
 
 #[derive(Debug)]
@@ -133,11 +137,27 @@ pub enum ScanIssueKind {
     NonUtf8Body,
 }
 
+#[derive(Debug)]
+pub struct ScanWarning {
+    pub path: PathBuf,
+    pub kind: ScanWarningKind,
+}
+
+#[derive(Debug, Error)]
+pub enum ScanWarningKind {
+    /// Top-level YAML was syntactically valid but not a mapping
+    /// (e.g., a stray top-level list). `serve` keeps the note with an
+    /// empty frontmatter; `doctor` reports it.
+    #[error("frontmatter root is not a YAML mapping")]
+    FrontmatterNotMapping,
+}
+
 /// Walk the vault and return every readable Markdown note alongside any
 /// per-file issues. The walk never panics on a single broken file.
 pub fn scan(vault_root: &Path) -> ScanResult {
     let mut notes = Vec::new();
     let mut issues = Vec::new();
+    let mut warnings = Vec::new();
 
     let walker = WalkDir::new(vault_root)
         .follow_links(false)
@@ -165,8 +185,16 @@ pub fn scan(vault_root: &Path) -> ScanResult {
         if !is_markdown(abs_path) {
             continue;
         }
-        match scan_one(vault_root, abs_path) {
-            Ok(note) => notes.push(note),
+        match scan_one_audit(vault_root, abs_path) {
+            Ok((note, maybe_warning)) => {
+                if let Some(kind) = maybe_warning {
+                    warnings.push(ScanWarning {
+                        path: abs_path.to_path_buf(),
+                        kind,
+                    });
+                }
+                notes.push(note);
+            }
             Err(kind) => issues.push(ScanIssue {
                 path: abs_path.to_path_buf(),
                 kind,
@@ -174,7 +202,11 @@ pub fn scan(vault_root: &Path) -> ScanResult {
         }
     }
 
-    ScanResult { notes, issues }
+    ScanResult {
+        notes,
+        issues,
+        warnings,
+    }
 }
 
 fn is_markdown(path: &Path) -> bool {
@@ -186,13 +218,28 @@ fn is_dot_prefixed(name: &std::ffi::OsStr) -> bool {
 }
 
 /// Read a single Markdown file off disk and produce its [`Note`]. Used by
-/// [`scan`] and by the filesystem watcher's per-event handler.
+/// the filesystem watcher's per-event handler -- which discards any
+/// soft-warning surface.
 ///
 /// # Errors
 /// Returns a [`ScanIssueKind`] when the file cannot be read, the body is not
 /// valid UTF-8, the path itself isn't UTF-8, or the frontmatter YAML fails to
 /// parse.
 pub fn scan_one(vault_root: &Path, abs_path: &Path) -> Result<Note, ScanIssueKind> {
+    scan_one_audit(vault_root, abs_path).map(|(note, _)| note)
+}
+
+/// Read a single Markdown file and surface both the [`Note`] and any
+/// soft warning (currently [`ScanWarningKind::FrontmatterNotMapping`]).
+/// Used by [`scan`] so `doctor` can report the diagnostic without
+/// changing what `serve` ingests.
+///
+/// # Errors
+/// Same as [`scan_one`].
+pub fn scan_one_audit(
+    vault_root: &Path,
+    abs_path: &Path,
+) -> Result<(Note, Option<ScanWarningKind>), ScanIssueKind> {
     let raw_bytes = std::fs::read(abs_path)?;
     let metadata = std::fs::metadata(abs_path)?;
     // Size from the bytes we actually hashed -- avoids the one-frame drift
@@ -204,7 +251,7 @@ pub fn scan_one(vault_root: &Path, abs_path: &Path) -> Result<Note, ScanIssueKin
 
     let text = std::str::from_utf8(&raw_bytes).map_err(|_| ScanIssueKind::NonUtf8Body)?;
     let (frontmatter_yaml, body) = split_frontmatter(text);
-    let frontmatter = parse_frontmatter(frontmatter_yaml)?;
+    let (frontmatter, warning) = parse_frontmatter_audit(frontmatter_yaml)?;
 
     let rel = abs_path
         .strip_prefix(vault_root)
@@ -214,7 +261,8 @@ pub fn scan_one(vault_root: &Path, abs_path: &Path) -> Result<Note, ScanIssueKin
     // Normalize separators for HTTP-facing storage; on Linux this is a no-op.
     let path = path.replace('\\', "/");
 
-    Ok(Note {
+    Ok((
+        Note {
             path,
             frontmatter,
             body: body.to_string(),
@@ -222,7 +270,9 @@ pub fn scan_one(vault_root: &Path, abs_path: &Path) -> Result<Note, ScanIssueKin
             last_modified,
             etag,
             size,
-    })
+        },
+        warning,
+    ))
 }
 
 /// Split a Markdown source into `(frontmatter_yaml, body)`. The frontmatter
@@ -250,15 +300,21 @@ fn split_frontmatter(src: &str) -> (&str, &str) {
     ("", src)
 }
 
-fn parse_frontmatter(yaml: &str) -> Result<Frontmatter, ScanIssueKind> {
+fn parse_frontmatter_audit(
+    yaml: &str,
+) -> Result<(Frontmatter, Option<ScanWarningKind>), ScanIssueKind> {
     if yaml.trim().is_empty() {
-        return Ok(BTreeMap::new());
+        return Ok((BTreeMap::new(), None));
     }
     let raw: serde_yaml::Value = serde_yaml::from_str(yaml)?;
     // A frontmatter that is not a mapping is not what Obsidian writes;
-    // treat as empty to avoid surfacing a contract surprise to consumers.
+    // serve keeps an empty frontmatter so the note is still served, but
+    // doctor sees the warning so a user can fix the file.
     let serde_yaml::Value::Mapping(map) = raw else {
-        return Ok(BTreeMap::new());
+        return Ok((
+            BTreeMap::new(),
+            Some(ScanWarningKind::FrontmatterNotMapping),
+        ));
     };
     let mut out = BTreeMap::new();
     for (k, v) in map {
@@ -268,7 +324,7 @@ fn parse_frontmatter(yaml: &str) -> Result<Frontmatter, ScanIssueKind> {
         };
         out.insert(key, coerce(v));
     }
-    Ok(out)
+    Ok((out, None))
 }
 
 fn coerce(v: serde_yaml::Value) -> Value {