dirsql 0.4.116__tar.gz → 0.4.117__tar.gz

{dirsql-0.4.116 → dirsql-0.4.117}/Cargo.lock

@@ -573,7 +573,7 @@ dependencies = [
 
 [[package]]
 name = "dirsql"
-version = "0.4.116"
+version = "0.4.117"
 dependencies = [
  "assert_cmd",
  "axum",
@@ -625,7 +625,7 @@ dependencies = [
 
 [[package]]
 name = "dirsql-py-ext"
-version = "0.4.116"
+version = "0.4.117"
 dependencies = [
  "dirsql",
  "pyo3",

{dirsql-0.4.116 → dirsql-0.4.117}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: dirsql
-Version: 0.4.116
+Version: 0.4.117
 Requires-Dist: bin-shim>=0.1
 Summary: Ephemeral SQL index over a local directory
 Keywords: sql,filesystem,directory,sqlite,index

{dirsql-0.4.116 → dirsql-0.4.117}/packages/python/Cargo.toml

@@ -4,7 +4,7 @@ name = "dirsql-py-ext"
 # pypi/maturin handler can rewrite it via `write-version` before
 # `maturin build`. `pyproject.toml` declares `dynamic = ["version"]`
 # and maturin reads this field. Mirrors `packages/rust/Cargo.toml`.
-version = "0.4.116"
+version = "0.4.117"
 # abi3 note: the `abi3-py310` pyo3 feature (below) builds ONE stable-ABI
 # `cp310-abi3` wheel per platform that loads on every CPython >= 3.10
 # (matching `pyproject.toml`'s `requires-python = ">=3.10"`), instead of

dirsql-0.4.117/packages/python/e2e-attestations/bot-1477-onelevel-speed-1016.json

@@ -0,0 +1,7 @@
+{
+  "command": "uv run python -m pytest tests/e2e/glob_speed_depth_bound_test.py tests/e2e/glob_speed_one_level_test.py tests/e2e/glob_speed_recursive_test.py tests/e2e/glob_speed_matchers_test.py tests/e2e/glob_speed_dotfiles_test.py tests/e2e/glob_speed_gitignore_test.py tests/e2e/glob_speed_node_modules_test.py tests/e2e/glob_speed_symlink_test.py tests/e2e/query_contract_test.py tests/e2e/fanout_test.py tests/e2e/use_case_join_test.py -q -s",
+  "ran_at": 1792409702,
+  "exit_code": 0,
+  "commit": "f500062c3a69da8d29e351d1a5580cff8f5af696",
+  "branch": "bot/1477-onelevel-speed-1016"
+}

dirsql-0.4.117/packages/python/e2e-attestations/bot-1524-path-eq-1016.json

@@ -0,0 +1,7 @@
+{
+  "command": "uv run python -m pytest tests/e2e/glob_speed_recursive_test.py tests/e2e/use_case_join_test.py -x -q",
+  "ran_at": 1792151597,
+  "exit_code": 0,
+  "commit": "05acaa67201b49a50f43b35cd3f3204d13a6a544",
+  "branch": "bot/1524-path-eq-1016"
+}

dirsql-0.4.117/packages/python/e2e-attestations/bot-1532-gitignore-fastest-bar-1016.json

@@ -0,0 +1,7 @@
+{
+  "command": "uv run python -m pytest tests/e2e/glob_speed_gitignore_test.py -x -q -s",
+  "ran_at": 1792155745,
+  "exit_code": 0,
+  "commit": "fb06f3cf1239d66c0cc2fff6a297448cbd12365a",
+  "branch": "bot/1532-gitignore-fastest-bar-1016"
+}

dirsql-0.4.117/packages/python/e2e-attestations/bot-1534-fm-native-1016.json

@@ -0,0 +1,7 @@
+{
+  "command": "uv run python -m pytest tests/e2e/frontmatter_parse_speed_test.py tests/e2e/jsonl_parse_speed_test.py tests/e2e/on_file_query_test.py tests/e2e/fanout_test.py tests/e2e/declared_table_name_test.py tests/e2e/query_contract_test.py tests/e2e/use_case_notebook_test.py tests/e2e/use_case_transcripts_test.py tests/e2e/sigint_interrupt_test.py -q -s",
+  "ran_at": 1792158472,
+  "exit_code": 0,
+  "commit": "4eaa7f0c26fa324fdb9b161e90fdb3fe22b8774f",
+  "branch": "bot/1534-fm-native-1016"
+}

{dirsql-0.4.116 → dirsql-0.4.117}/packages/rust/Cargo.toml

@@ -6,7 +6,7 @@ description = "Ephemeral SQL index over a local directory"
 # version stays for the binding crates (publish = false). The literal
 # intentionally lags the published version: putitoutthere rewrites it to
 # the planned release version at build time and never commits it back.
-version = "0.4.116"
+version = "0.4.117"
 edition.workspace = true
 # Literal `license` (not `license.workspace = true`) because
 # putitoutthere's preflight check reads `[package].license` directly

{dirsql-0.4.116 → dirsql-0.4.117}/packages/rust/src/db.rs

@@ -312,6 +312,63 @@ impl ShapedRows {
     }
 }
 
+/// [`Db::shape_rows`] against `columns`, the table's columns in DDL order,
+/// so rows can be shaped off the connection's thread.
+pub(crate) fn shape_json_rows(
+    table: &str,
+    columns: Vec<String>,
+    rows: Vec<JsonRow>,
+    strict: bool,
+) -> Result<ShapedRows> {
+    let mut cells = Vec::with_capacity(rows.len());
+    for row in rows {
+        if strict {
+            for (key, _) in &row.0 {
+                validate_identifier(key)?;
+            }
+            let extra: Vec<&str> = row
+                .0
+                .iter()
+                .map(|(key, _)| &**key)
+                .filter(|key| !columns.iter().any(|c| c == key))
+                .collect();
+            if !extra.is_empty() {
+                return Err(DbError::SchemaMismatch(format!(
+                    "extra columns not in table {}: {}",
+                    table,
+                    extra.join(", ")
+                )));
+            }
+            let missing: Vec<&str> = columns
+                .iter()
+                .map(String::as_str)
+                .filter(|c| !row.0.iter().any(|(key, _)| **key == **c))
+                .collect();
+            if !missing.is_empty() {
+                return Err(DbError::SchemaMismatch(format!(
+                    "missing columns for table {}: {}",
+                    table,
+                    missing.join(", ")
+                )));
+            }
+        }
+        let in_order = row.0.len() == columns.len()
+            && row.0.iter().zip(&columns).all(|((key, _), c)| **key == **c);
+        if in_order {
+            cells.push(row.0.into_iter().map(|(_, v)| json_into_value(v)).collect());
+            continue;
+        }
+        let mut shaped = vec![Value::Null; columns.len()];
+        for (key, value) in row.0 {
+            if let Some(i) = columns.iter().position(|c| **c == *key) {
+                shaped[i] = json_into_value(value);
+            }
+        }
+        cells.push(shaped);
+    }
+    Ok(ShapedRows { columns, cells })
+}
+
 fn json_into_value(value: serde_json::Value) -> Value {
     match value {
         serde_json::Value::String(s) => Value::Text(s),
@@ -566,9 +623,10 @@ impl Db {
     /// Return the column names declared in `table`'s DDL.
     pub fn get_table_columns(&self, table: &str) -> Result<Vec<String>> {
         validate_identifier(table)?;
-        let mut stmt = self
-            .conn
-            .prepare(&format!("PRAGMA table_info({})", table))?;
+        let mut stmt = self.conn.prepare(&format!(
+            "PRAGMA table_info({})",
+            crate::sql_literal::quote_identifier(table)
+        ))?;
         let columns: Vec<String> = stmt
             .query_map([], |row| row.get::<_, String>(1))?
             .filter_map(|r| r.ok())
@@ -714,54 +772,7 @@ impl Db {
     /// a less-actionable "extra columns" mismatch. Every row is shaped before
     /// any is returned, so a hook that got one row wrong fails as a unit.
     pub fn shape_rows(&self, table: &str, rows: Vec<JsonRow>, strict: bool) -> Result<ShapedRows> {
-        let columns = self.get_table_columns(table)?;
-        let mut cells = Vec::with_capacity(rows.len());
-        for row in rows {
-            if strict {
-                for (key, _) in &row.0 {
-                    validate_identifier(key)?;
-                }
-                let extra: Vec<&str> = row
-                    .0
-                    .iter()
-                    .map(|(key, _)| &**key)
-                    .filter(|key| !columns.iter().any(|c| c == key))
-                    .collect();
-                if !extra.is_empty() {
-                    return Err(DbError::SchemaMismatch(format!(
-                        "extra columns not in table {}: {}",
-                        table,
-                        extra.join(", ")
-                    )));
-                }
-                let missing: Vec<&str> = columns
-                    .iter()
-                    .map(String::as_str)
-                    .filter(|c| !row.0.iter().any(|(key, _)| **key == **c))
-                    .collect();
-                if !missing.is_empty() {
-                    return Err(DbError::SchemaMismatch(format!(
-                        "missing columns for table {}: {}",
-                        table,
-                        missing.join(", ")
-                    )));
-                }
-            }
-            let in_order = row.0.len() == columns.len()
-                && row.0.iter().zip(&columns).all(|((key, _), c)| **key == **c);
-            if in_order {
-                cells.push(row.0.into_iter().map(|(_, v)| json_into_value(v)).collect());
-                continue;
-            }
-            let mut shaped = vec![Value::Null; columns.len()];
-            for (key, value) in row.0 {
-                if let Some(i) = columns.iter().position(|c| **c == *key) {
-                    shaped[i] = json_into_value(value);
-                }
-            }
-            cells.push(shaped);
-        }
-        Ok(ShapedRows { columns, cells })
+        shape_json_rows(table, self.get_table_columns(table)?, rows, strict)
     }
 
     /// Insert every row of `rows` under `file_path`, in order, through one
@@ -2476,6 +2487,15 @@ mod tests {
         assert!(result.is_err());
     }
 
+    #[test]
+    fn get_table_columns_reads_a_table_named_by_a_keyword() {
+        let db = Db::new().unwrap();
+        db.create_table("order", "CREATE TABLE \"order\" (name TEXT)")
+            .unwrap();
+        let cols = db.get_table_columns("order").unwrap();
+        assert!(cols.contains(&"name".to_string()), "{cols:?}");
+    }
+
     #[test]
     fn get_table_columns_nonexistent_table_returns_empty() {
         let db = Db::new().unwrap();

{dirsql-0.4.116 → dirsql-0.4.117}/packages/rust/src/lib.rs

@@ -1288,20 +1288,11 @@ impl DirSQL {
         // (and that pass the racy-window check) are trusted instead of
         // re-parsed.
         let (scanned_files, deleted) = match &persist_ctx {
-            None => {
-                let files = scanned
-                    .into_iter()
-                    .map(|found| ScannedFile {
-                        abs_path: found.path,
-                        rel_path: found.rel_path,
-                        table_name: found.table_name,
-                        stat: None,
-                        trusted: false,
-                    })
-                    .collect();
-                (files, Vec::new())
+            None => (Scanned::Found(scanned), Vec::new()),
+            Some(ctx) => {
+                let (files, deleted) = reconcile_scan(scanned, ctx, &RealFs)?;
+                (Scanned::Files(files), deleted)
             }
-            Some(ctx) => reconcile_scan(scanned, ctx, &RealFs)?,
         };
 
         Ok(PreparedBuild {
@@ -1482,17 +1473,22 @@ impl DirSQL {
         let mut batch_files: HashMap<String, Vec<String>> = HashMap::new();
 
         // Grouping walks every scanned file, so it runs beside the early
-        // hooks' inserts instead of ahead of them.
+        // hooks' inserts instead of ahead of them, and settles the early
+        // tables' file lists there too.
+        let early_tables: HashSet<String> = early.keys().cloned().collect();
         let (grouped, streamed) = std::thread::scope(|scope| {
-            let grouping = scope.spawn(|| group_scanned(scanned_files, &on_file_map));
+            let grouping = scope.spawn(|| {
+                group_scanned(scanned_files.into_files(), &on_file_map)
+                    .map(|(per_file, batches)| (per_file, split_streamed(batches, &early_tables)))
+            });
             let streamed = stream_early(&db, early, &strict_map);
             (
                 grouping.join().expect("grouping scanned files panicked"),
                 streamed,
             )
         });
-        let streamed = streamed?;
-        let (per_file, batches) = grouped?;
+        streamed?;
+        let (per_file, (batches, streamed)) = grouped?;
 
         // The phase a user actually waits on: one `on_file` round trip per
         // file or per table, plus whatever the table's DDL triggers on insert.
@@ -1501,8 +1497,15 @@ impl DirSQL {
             + batches
                 .iter()
                 .map(|(_, _, files)| files.len())
-                .sum::<usize>()) as u64;
+                .sum::<usize>()
+            + streamed.iter().map(|(_, rel)| rel.len()).sum::<usize>())
+            as u64;
         let mut done: u64 = 0;
+        for (table_name, rel_paths) in streamed {
+            progress.update(done, Some(total_files));
+            done += rel_paths.len() as u64;
+            batch_files.insert(table_name, rel_paths);
+        }
         for (
             ScannedFile {
                 rel_path,
@@ -1580,7 +1583,7 @@ impl DirSQL {
             let current = persist_ready.as_ref().is_some_and(|(deleted, _, _)| {
                 files.iter().all(|f| f.trusted) && !deleted.iter().any(|(_, t)| *t == table_name)
             });
-            if !current && !streamed.contains(&table_name) {
+            if !current {
                 let table_root = anchors.of(&table_name);
                 let production = start_hook(hook, abs_paths.into_iter());
                 let outcome = stream_table_rows(&db, &table_name, production, strict)
@@ -2004,6 +2007,33 @@ pub struct ScannedFile {
     pub trusted: bool,
 }
 
+/// The files the startup scan found: as the walk found them, or already
+/// reconciled against the persistent cache.
+enum Scanned {
+    Found(Vec<Found>),
+    Files(Vec<ScannedFile>),
+}
+
+impl Scanned {
+    /// Every file as a [`ScannedFile`]. A fresh walk's are made here rather
+    /// than in prepare, so the copy runs off the thread that inserts rows.
+    fn into_files(self) -> Vec<ScannedFile> {
+        match self {
+            Self::Found(found) => found
+                .into_iter()
+                .map(|found| ScannedFile {
+                    abs_path: found.path,
+                    rel_path: found.rel_path,
+                    table_name: found.table_name,
+                    stat: None,
+                    trusted: false,
+                })
+                .collect(),
+            Self::Files(files) => files,
+        }
+    }
+}
+
 /// Intermediate state produced by [`DirSQL::prepare_resolved`] and consumed
 /// by [`DirSQL::finish_build`]. Opaque on purpose; fields are only visible
 /// to the in-workspace bindings.
@@ -2020,7 +2050,7 @@ pub struct PreparedBuild {
     /// The configured skip rules, carried through so path-table scans apply
     /// the same ones declared tables do.
     ignore: Vec<String>,
-    scanned_files: Vec<ScannedFile>,
+    scanned_files: Scanned,
     /// Per-table hooks already running over the paths the scan found, by table.
     early: HashMap<String, Production>,
     persist: Option<PreparedPersist>,
@@ -2437,9 +2467,10 @@ fn collect_table_rows(
 
 /// Replace a per-table hook's rows under [`BATCH_OWNER`], storing each chunk
 /// the hook emits while the hook, on its own thread, goes on producing the
-/// next. The caller's open transaction keeps a failure from being seen: the
-/// outer `Err` is SQLite's, the inner one the hook's or a row's. A hook
-/// failure outranks a row that failed to shape.
+/// next. Chunks are shaped on a thread of their own, in order, so shaping one
+/// overlaps inserting the last. The caller's open transaction keeps a failure
+/// from being seen: the outer `Err` is SQLite's, the inner one the hook's or a
+/// row's. A hook failure outranks a row that failed to shape.
 fn stream_table_rows(
     db: &Db,
     table: &str,
@@ -2447,17 +2478,32 @@ fn stream_table_rows(
     strict: bool,
 ) -> db::Result<std::result::Result<(), String>> {
     db.delete_rows_by_file(table, BATCH_OWNER)?;
-    let mut shaped = Ok(());
-    for chunk in production.chunks {
-        if shaped.is_err() {
-            continue;
+    let columns = db.get_table_columns(table)?;
+    let Production { chunks, handle } = production;
+    let (shaped_tx, shaped_rx) = std::sync::mpsc::channel();
+    let shaper = std::thread::spawn({
+        let table = table.to_string();
+        move || {
+            for chunk in chunks {
+                let shaped = db::shape_json_rows(&table, columns.clone(), chunk, strict);
+                if shaped_tx.send(shaped).is_err() {
+                    return;
+                }
+            }
         }
-        match db.shape_rows(table, chunk, strict) {
+    });
+    let mut shaped = Ok(());
+    for rows in shaped_rx {
+        match rows {
             Ok(rows) => db.insert_rows(table, &rows, BATCH_OWNER)?,
-            Err(e) => shaped = Err(e.to_string()),
+            Err(e) => {
+                shaped = Err(e.to_string());
+                break;
+            }
         }
     }
-    let produced = production.handle.join().expect("per-table hook panicked");
+    shaper.join().expect("shaping a hook's rows panicked");
+    let produced = handle.join().expect("per-table hook panicked");
     Ok(produced.and(shaped))
 }
 
@@ -2591,14 +2637,12 @@ fn group_scanned(
     Ok((per_file, batches))
 }
 
-/// Insert the rows of every hook started before the scan, returning the
-/// tables streamed.
+/// Insert the rows of every hook started before the scan.
 fn stream_early(
     db: &Db,
     early: HashMap<String, Production>,
     strict_map: &HashMap<String, bool>,
-) -> Result<HashSet<String>> {
-    let mut streamed = HashSet::new();
+) -> Result<()> {
     for (table_name, production) in early {
         let strict = *strict_map.get(&table_name).unwrap_or(&false);
         if let Err(message) =
@@ -2609,9 +2653,30 @@ fn stream_early(
                 message,
             });
         }
-        streamed.insert(table_name);
     }
-    Ok(streamed)
+    Ok(())
+}
+
+/// Split off the batches of tables whose hooks already ran, keeping only
+/// each one's relative paths in scan order.
+#[allow(clippy::type_complexity)]
+fn split_streamed(
+    batches: Vec<(String, Arc<OnFilesFn>, Vec<ScannedFile>)>,
+    streamed: &HashSet<String>,
+) -> (
+    Vec<(String, Arc<OnFilesFn>, Vec<ScannedFile>)>,
+    Vec<(String, Vec<String>)>,
+) {
+    let mut rest = Vec::new();
+    let mut done = Vec::new();
+    for (table_name, hook, mut files) in batches {
+        if streamed.contains(&table_name) {
+            done.push((table_name, batch_paths(&mut files, false).0));
+        } else {
+            rest.push((table_name, hook, files));
+        }
+    }
+    (rest, done)
 }
 
 fn batch_paths(files: &mut [ScannedFile], keep_rel: bool) -> (Vec<String>, Vec<PathBuf>) {
@@ -3154,7 +3219,7 @@ mod internal_tests {
             0,
             "zero tables can match nothing, so the startup walk is pure waste"
         );
-        assert!(prepared.scanned_files.is_empty());
+        assert!(prepared.scanned_files.into_files().is_empty());
     }
 
     /// The skip is narrow: one declared table still walks, and prepare still
@@ -3175,10 +3240,11 @@ mod internal_tests {
         .unwrap();
 
         assert_eq!(scan.calls.get(), 1);
-        let found: Vec<&str> = prepared
+        let found: Vec<String> = prepared
             .scanned_files
-            .iter()
-            .map(|f| f.rel_path.as_str())
+            .into_files()
+            .into_iter()
+            .map(|f| f.rel_path)
             .collect();
         assert_eq!(found, vec!["a.txt"]);
     }
@@ -3295,13 +3361,13 @@ mod internal_tests {
             functions: Vec::new(),
             groups: Vec::new(),
             anchors: TableAnchors::at_root(dir.path()),
-            scanned_files: vec![ScannedFile {
+            scanned_files: Scanned::Files(vec![ScannedFile {
                 abs_path: PathBuf::new(),
                 rel_path: "ghost.txt".into(),
                 table_name: "ghost".into(),
                 stat: None,
                 trusted: false,
-            }],
+            }]),
             early: HashMap::new(),
             poll_interval: DEFAULT_POLL_INTERVAL,
             persist: None,
@@ -5367,13 +5433,15 @@ mod internal_tests {
         scanned_files: Vec<ScannedFile>,
         persist: Option<PreparedPersist>,
     ) -> PreparedBuild {
-        let scanned_files = scanned_files
-            .into_iter()
-            .map(|mut file| {
-                file.abs_path = root.join(&file.rel_path);
-                file
-            })
-            .collect();
+        let scanned_files = Scanned::Files(
+            scanned_files
+                .into_iter()
+                .map(|mut file| {
+                    file.abs_path = root.join(&file.rel_path);
+                    file
+                })
+                .collect(),
+        );
         PreparedBuild {
             ignore: Vec::new(),
             root: root.to_path_buf(),
@@ -5710,6 +5778,48 @@ mod internal_tests {
         }
     }
 
+    #[test]
+    fn finish_build_counts_the_files_of_a_hook_started_before_the_scan_toward_the_ingest_total() {
+        let dir = TempDir::new().unwrap();
+        let out = Captured::default();
+        let progress = Progress::new(
+            "indexing",
+            "indexed",
+            "files",
+            Box::new(out.clone()),
+            Box::new(progress::SystemClock),
+            progress::Mode::Always,
+            false,
+        );
+        let notes = Table::new("notes", "CREATE TABLE notes (name TEXT)", "*.md", |_| {
+            Vec::new()
+        });
+        let (items, _) = recording_per_table("items", "*.txt");
+        let (feeds, early) = start_per_table_hooks(std::slice::from_ref(&items));
+        for name in ["a.txt", "b.txt"] {
+            feeds["items"].send(dir.path().join(name)).unwrap();
+        }
+        drop(feeds);
+        let scanned = vec![
+            scanned("a.txt", "items", false),
+            scanned("b.txt", "items", false),
+            scanned("n.md", "notes", false),
+        ];
+        let mut prepared = prepared_build(dir.path(), vec![notes, items], scanned, None);
+        prepared.early = early;
+
+        DirSQL::finish_build_with_fs(prepared, Arc::new(FakeFs::default()), progress).unwrap();
+
+        let text = out.text();
+        for line in [
+            "dirsql: indexing 0/3 files (0%)",
+            "dirsql: indexing 2/3 files (66%)",
+            "dirsql: indexed 3 files in",
+        ] {
+            assert!(text.contains(line), "missing {line:?} in {text:?}");
+        }
+    }
+
     #[test]
     fn finish_build_skips_a_per_file_hook_for_a_file_the_cache_holds() {
         let dir = TempDir::new().unwrap();