fix(prove): trim non-subtree keys from both ends of the sorted key slice

src/db.rs

@@ -12,7 +12,7 @@ use std::{
     sync::{Arc, Mutex},
 };
 
-const HEADER_MAGIC: [u8; 9] = [b's', b'p', b'a', b'c', b'e', b':', b'/', b'/', b'.'];
+const HEADER_MAGIC: [u8; 9] = *b"space://.";
 pub(crate) const CHUNK_SIZE: u64 = 4096;
 pub(crate) const HEADER_SIZE: u64 = CHUNK_SIZE * 2;
 

src/tx.rs

@@ -518,9 +518,18 @@ impl<H: NodeHasher> ReadTransaction<H> {
                 left,
                 right,
             } => {
-                // Exclude keys that are not in this subtree.
-                let end = keys.partition_point(|key| key.split_point(depth, *prefix).is_none());
-                let keys = &keys[..end];
+                // Exclude keys that are not in this subtree. Sorted keys that
+                // diverge below the prefix come before the matching band, ones
+                // that diverge above come after — trim both sides.
+                let start = keys.partition_point(|key| {
+                    key.split_point(depth, *prefix)
+                        .is_some_and(|p| key.direction(depth + p) == Direction::Left)
+                });
+                let end = keys.partition_point(|key| {
+                    key.split_point(depth, *prefix)
+                        .is_none_or(|p| key.direction(depth + p) == Direction::Left)
+                });
+                let keys = &keys[start..end];
 
                 // Keys are split based on their direction at the current depth.
                 let depth = depth + prefix.bit_len();
@@ -611,6 +620,9 @@ impl<H: NodeHasher> ReadTransaction<H> {
         }
     }
 
+    // `hash_index` is read directly only under the `hash-idx` feature;
+    // without it, clippy sees a recursion-only parameter.
+    #[cfg_attr(not(feature = "hash-idx"), allow(clippy::only_used_in_recursion))]
     fn hash_node<'c>(
         db: &Database<H>,
         cache: &mut Cache,

tests/batch_insert_proof.rs

@@ -0,0 +1,86 @@
+//! `prove(keys, Standard)` should return a subtree carrying enough of the tree
+//! to insert those same keys.
+use spacedb::db::Database;
+use spacedb::subtree::ValueOrHash;
+use spacedb::tx::ProofType;
+use spacedb::{Hash, NodeHasher, Sha256Hasher};
+
+fn key(tag: &str, i: usize) -> Hash {
+    Sha256Hasher::hash(format!("{}{}", tag, i).as_bytes())
+}
+
+/// In-memory tree pre-populated with `existing` keys.
+fn tree_with(existing: usize) -> Database<Sha256Hasher> {
+    let db = Database::memory().expect("memory db");
+    let mut tx = db.begin_write().expect("write tx");
+    for i in 0..existing {
+        tx = tx
+            .insert(key("existing", i), b"v".to_vec())
+            .expect("insert");
+    }
+    tx.commit().expect("commit");
+    db
+}
+
+/// Prove `n` absent keys, then insert exactly those keys into the returned
+/// subtree. Returns the index of the first insert that failed.
+fn replay(db: &Database<Sha256Hasher>, n: usize) -> Option<usize> {
+    let keys: Vec<Hash> = (0..n).map(|i| key("new", i)).collect();
+
+    let mut snapshot = db.begin_read().expect("read tx");
+    let mut proof = snapshot
+        .prove(&keys, ProofType::Standard)
+        .expect("prove should succeed");
+
+    for (i, k) in keys.iter().enumerate() {
+        if proof.insert(*k, ValueOrHash::Hash(key("val", i))).is_err() {
+            return Some(i);
+        }
+    }
+    None
+}
+
+/// A proof over N keys should support inserting those N keys.
+#[test]
+fn standard_proof_supports_inserting_its_own_keys() {
+    for existing in [2usize, 102, 1000] {
+        let db = tree_with(existing);
+        assert_eq!(
+            replay(&db, 900),
+            None,
+            "tree with {} existing keys: proving 900 absent keys returned a \
+             subtree that cannot insert all 900",
+            existing
+        );
+    }
+}
+
+/// Proving a superset of keys must not support fewer insertions than proving
+/// a subset — the larger proof strictly contains more of the tree.
+#[test]
+fn proving_more_keys_does_not_reduce_capability() {
+    let db = tree_with(102);
+
+    // Largest batch that replays cleanly.
+    let mut largest = 0;
+    for n in 1..=512 {
+        if replay(&db, n).is_none() {
+            largest = n;
+        } else {
+            break;
+        }
+    }
+    assert!(largest > 0, "no batch size replayed at all");
+
+    // A proof over more keys must still handle at least that many inserts.
+    let failed_at = replay(&db, largest * 2);
+    assert!(
+        failed_at.is_none() || failed_at.unwrap() >= largest,
+        "proving {} keys supports {} inserts, but proving {} keys fails at \
+         entry {} — a larger proof supports fewer insertions",
+        largest,
+        largest,
+        largest * 2,
+        failed_at.unwrap()
+    );
+}