refactor: remove capacity in Compactor::build_tables (#443)

Author: ethe

guide/src/contribution/composite_primary_keys.md

@@ -60,6 +60,7 @@ This document outlines a practical, incremental plan to add composite (multi-col
 
 - Extend `DynSchema` to track `primary_indices: Vec<usize>` in metadata (replacing the single `primary_key_index`).
 - Update `DynRecordRef::new` and readers to honor multiple PK indices.
+- Ensure tombstone writes (row == None) still populate all PK columns from `Ts.key` so ordering/lookups remain correct.
 - Define a composite key wrapper for `Value`/`ValueRef` (or generate a per-dyn-schema composite type if feasible). Initially out-of-scope for phase 1.
 
 ## Step-by-Step Plan
@@ -103,15 +104,16 @@ Phase 3: Dynamic records (optional)
 7. `DynSchema` multi-PK
    - Store `primary_indices` metadata; update dynamic arrays/refs to keep all PK columns in projections.
    - Provide a composite `ValueRef` key wrapper for in-memory operations.
-   - Acceptance: dynamic tests mirroring integration scenarios pass.
+   - Ensure tombstones populate PK components from `Ts.key` in builders (e.g., `DynRecordBuilder::push`).
+   - Acceptance: dynamic tests mirroring integration scenarios pass, including tombstone rows retaining all PK components.
 
 ## Code Touchpoints
 
 - Traits/APIs: `src/record/mod.rs` (Schema), `src/option.rs` (DbOption::new)
 - Read paths: `src/lib.rs` (get/scan/package), `src/transaction.rs` (get/scan)
 - Macro codegen: `tonbo_macros/src/record.rs`, `tonbo_macros/src/keys.rs`, `tonbo_macros/src/data_type.rs`
 - Key types: `src/record/key/composite/`
-- Dynamic (phase 3): `src/record/dynamic/*`
+- Dynamic (phase 3): `src/record/dynamic/*` (incl. tombstone handling in `DynRecordBuilder::push`)
 
 ## Testing Strategy
 
@@ -126,6 +128,9 @@ Phase 3: Dynamic records (optional)
   - WAL/compaction unaffected (basic smoke tests).
 - (Optional) Property tests: ordering equivalence vs. native tuple lexicographic ordering when Option B is used.
 
+- Tombstones:
+  - For both single- and multi-column PKs, verify that tombstone rows (row == None) keep all PK column values populated from `Ts.key` in Arrow arrays and through `RecordRef::from_record_batch`.
+
 ## Backward Compatibility & Migration
 
 - All existing single-PK code continues to work without changes due to default-impl fallbacks.
@@ -167,3 +172,4 @@ pub struct User {
 - [ ] Configure Parquet sorting/statistics for all PK components.
 - [ ] Add unit/trybuild/integration tests.
 - [ ] Update user guide (mention composite PK support and examples).
+ - [ ] Ensure tombstone rows keep PK component values (builders/readers), validated by tests.

src/compaction/mod.rs

@@ -59,9 +59,8 @@ where
     ) -> Result<(), CompactionError<R>> {
         let mut stream = MergeStream::<R>::from_vec(streams, u32::MAX.into(), None).await?;
 
-        // Kould: is the capacity parameter necessary?
         let mut builder =
-            <R::Schema as RecordSchema>::Columns::builder(schema.arrow_schema().clone(), 8192);
+            <R::Schema as RecordSchema>::Columns::builder(schema.arrow_schema().clone(), 0);
         let mut min = None;
         let mut max = None;
 

src/stream/level.rs

@@ -66,7 +66,6 @@ impl<'level, R> LevelStream<'level, R>
 where
     R: Record,
 {
-    // Kould: only used by Compaction now, and the start and end of the sstables range are known
     #[allow(clippy::too_many_arguments)]
     pub(crate) fn new(
         version: &Version<R>,
@@ -80,6 +79,7 @@ where
         ts: Timestamp,
         limit: Option<usize>,
         projection_mask: ProjectionMask,
+        // TODO: Refactor some top level components to a context structure.
         fs: Arc<dyn DynFs>,
         parquet_lru: Arc<dyn DynLruCache<Ulid> + Send + Sync>,
         order: Option<Order>,