feat(p10-1a-2): backfill SearchHit.repo from doc metadata (Task 8b)

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

crates/kebab-app/src/app.rs

@@ -40,8 +40,8 @@ use anyhow::{Context, Result, anyhow};
 use lru::LruCache;
 
 use kebab_core::{
-    Answer, Embedder, IndexVersion, LanguageModel, Retriever, SearchHit, SearchMode,
-    SearchOpts, SearchQuery, VectorStore,
+    Answer, DocumentStore, Embedder, IndexVersion, LanguageModel, Retriever, SearchHit,
+    SearchMode, SearchOpts, SearchQuery, VectorStore,
 };
 use kebab_embed_local::FastembedEmbedder;
 use kebab_llm_local::OllamaLanguageModel;
@@ -301,6 +301,10 @@ impl App {
         // `code_lang: None`; we own the post-processing here in kebab-app
         // and can fill it cheaply from data already present in the hit.
         backfill_code_lang(&mut hits);
+        // p10-1A-2 Task 8b: backfill `repo` from the document's
+        // `Metadata.repo`. Unlike `code_lang`, this cannot be derived from
+        // the Citation alone — it requires a store lookup by `doc_id`.
+        self.backfill_repo(&mut hits);
         Ok(hits)
     }
 
@@ -394,6 +398,8 @@ impl App {
             );
             // p10-1A-2: backfill code_lang — same as search_uncached.
             backfill_code_lang(&mut traced_hits);
+            // p10-1A-2 Task 8b: backfill repo — same as search_uncached.
+            self.backfill_repo(&mut traced_hits);
 
             // Apply offset + k_effective truncation (mirrors non-trace path).
             let drop_n = offset.min(traced_hits.len());
@@ -784,6 +790,47 @@ impl App {
         }
     }
 
+    /// p10-1A-2 Task 8b: back-fill `SearchHit.repo` from the originating
+    /// document's `Metadata.repo` for every hit whose `repo` field is
+    /// currently `None`. The search layer (kebab-search) constructs hits
+    /// with `repo: None` because it has no store access; we fill it here
+    /// in kebab-app post-retrieval via a per-distinct-`doc_id` store lookup.
+    ///
+    /// Deduplication: a small `HashMap` accumulates the
+    /// `(doc_id → Option<String>)` mapping so each unique document is
+    /// fetched at most once. Search result sets are small (default k ≤ 20),
+    /// so the map overhead is negligible. A `None` entry is cached too
+    /// (document not found or no repo in metadata) to avoid re-querying.
+    ///
+    /// Non-repo documents (markdown, PDF, plain text, code files outside a
+    /// git tree) correctly keep `repo: None` — `Metadata.repo` is already
+    /// `None` for those, so the assignment is a no-op.
+    fn backfill_repo(&self, hits: &mut [SearchHit]) {
+        use std::collections::HashMap;
+        use kebab_core::DocumentId;
+
+        // doc_id → Option<String> where None means "not found / no repo"
+        let mut cache: HashMap<DocumentId, Option<String>> = HashMap::new();
+
+        for hit in hits.iter_mut() {
+            if hit.repo.is_some() {
+                continue;
+            }
+            let repo_val = cache
+                .entry(hit.doc_id.clone())
+                .or_insert_with(|| {
+                    self.sqlite
+                        .get_document(&hit.doc_id)
+                        .ok()
+                        .flatten()
+                        .and_then(|doc| doc.metadata.repo)
+                });
+            if let Some(r) = repo_val {
+                hit.repo = Some(r.clone());
+            }
+        }
+    }
+
     /// Resolve the embedder + vector store, surfacing the user-friendly
     /// "switch to --mode lexical" error when embeddings are disabled.
     fn require_embeddings(

crates/kebab-app/tests/code_ingest_smoke.rs

@@ -101,6 +101,64 @@ fn rust_file_ingests_and_searches_as_code_citation() {
     );
 }
 
+/// p10-1A-2 Task 8b: a code search hit must carry `SearchHit.repo` filled
+/// from the document's `Metadata.repo` (which is set by `detect_repo` during
+/// ingest). `detect_repo` returns the name of the directory that contains
+/// `.git/`, so we `git init` the workspace root before ingesting and then
+/// assert that `h.repo == Some("workspace")`.
+#[test]
+fn rust_code_search_hit_has_repo() {
+    let env = TestEnv::lexical_only();
+
+    // `detect_repo` walks up from the file looking for `.git/`.
+    // Initialise a bare git repo at the workspace root so it is
+    // discoverable. We only need the `.git/` directory — no commits
+    // required.
+    let git_status = std::process::Command::new("git")
+        .args(["init", "--quiet"])
+        .arg(env.workspace_root.as_os_str())
+        .status()
+        .expect("git init");
+    assert!(git_status.success(), "git init must succeed");
+
+    std::fs::write(
+        env.workspace_root.join("repo_demo.rs"),
+        "/// multiplies two integers\npub fn mul(a: i32, b: i32) -> i32 {\n    a * b\n}\n",
+    )
+    .unwrap();
+
+    let report =
+        kebab_app::ingest_with_config(env.config.clone(), env.scope(), false)
+            .expect("ingest must succeed");
+    assert_eq!(report.errors, 0, "no ingest errors: {report:?}");
+
+    let hits = kebab_app::search_with_config(env.config.clone(), lexical_query("mul"))
+        .expect("search must succeed");
+
+    let h = hits
+        .iter()
+        .find(|h| matches!(&h.citation, Citation::Code { .. }))
+        .expect("at least one Citation::Code hit for 'mul'");
+
+    // The workspace root directory is named "workspace" by `TestEnv`.
+    let expected_repo = env
+        .workspace_root
+        .file_name()
+        .and_then(|n| n.to_str())
+        .map(str::to_owned);
+    assert_eq!(
+        h.repo,
+        expected_repo,
+        "SearchHit.repo must match the workspace dir name (detect_repo result)"
+    );
+    // Also sanity-check code_lang is still filled.
+    assert_eq!(
+        h.code_lang.as_deref(),
+        Some("rust"),
+        "SearchHit.code_lang must be 'rust'"
+    );
+}
+
 /// Re-ingesting the same `.rs` file without changes must report
 /// `Unchanged` (incremental-skip path exercised).
 #[test]