fix: resolve system index types in describe_indices (#6685)

describe_indices() returned "Unknown" for system indices
(__lance_frag_reuse, __lance_mem_wal) because their proto details have
no scalar plugin registered. list_indices() already special-cases these
via infer_system_index_type(); apply the same check in
IndexDescriptionImpl::try_new so the two methods agree.

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

Author: wkalt

python/python/tests/test_optimize.py

@@ -9,6 +9,7 @@
 import lance
 import numpy as np
 import pyarrow as pa
+import pytest
 from lance.lance import Compaction
 from lance.optimize import RewriteResult
 from lance.vector import vec_to_table
@@ -323,6 +324,41 @@ def test_defer_index_remap(tmp_path: Path):
     assert any(idx.name == "__lance_frag_reuse" for idx in indices)
 
 
+@pytest.mark.filterwarnings("ignore::DeprecationWarning")
+def test_describe_indices_matches_list_indices_for_frag_reuse(tmp_path: Path):
+    """describe_indices() and list_indices() must agree on the index_type
+    string for every index, including the __lance_frag_reuse system index
+    that defer_index_remap produces.
+
+    list_indices() special-cases system indices via infer_system_index_type()
+    in python/src/dataset.rs. describe_indices() in
+    rust/lance/src/index.rs::IndexDescriptionImpl::try_new does not, so it
+    falls through to a plugin lookup that has no entry for
+    FragmentReuseIndexDetails and reports 'Unknown' instead.
+    """
+    base_dir = tmp_path / "dataset"
+    data = pa.table({"i": range(6_000), "val": range(6_000)})
+    dataset = lance.write_dataset(data, base_dir, max_rows_per_file=1_000)
+    dataset.create_scalar_index("i", "BTREE")
+    dataset.delete("i < 500")
+    dataset.optimize.compact_files(
+        target_rows_per_fragment=2_000, defer_index_remap=True, num_threads=1
+    )
+
+    dataset = lance.dataset(base_dir)
+    described = {d.name: d.index_type for d in dataset.describe_indices()}
+    listed = {idx["name"]: idx["type"] for idx in dataset.list_indices()}
+
+    assert "__lance_frag_reuse" in listed, (
+        "test precondition: defer_index_remap should produce a frag-reuse index"
+    )
+    assert described == listed, (
+        "describe_indices and list_indices disagree on index_type:\n"
+        f"  describe_indices: {described}\n"
+        f"  list_indices:     {listed}"
+    )
+
+
 def test_dataset_distributed_optimize(tmp_path: Path):
     base_dir = tmp_path / "dataset"
     data = pa.table({"a": range(800), "b": range(800)})

rust/lance/src/index.rs

@@ -46,7 +46,7 @@ use lance_index::vector::sq::ScalarQuantizer;
 use lance_index::vector::v3::subindex::IvfSubIndex;
 use lance_index::{INDEX_FILE_NAME, Index, IndexType, PrewarmOptions, pb, vector::VectorIndex};
 use lance_index::{
-    IndexCriteria, is_system_index,
+    IndexCriteria, infer_system_index_type, is_system_index,
     metrics::{MetricsCollector, NoOpMetricsCollector},
 };
 use lance_io::scheduler::{ScanScheduler, SchedulerConfig};
@@ -651,8 +651,14 @@ impl IndexDescriptionImpl {
         let details = IndexDetails(index_details.clone());
         let mut rows_indexed = 0;
 
-        // Vector indices need to be opened to get the correct type
-        let index_type = if details.is_vector() {
+        // System indices (e.g. __lance_frag_reuse, __lance_mem_wal) are
+        // identified by name and have no entry in the scalar plugin registry,
+        // so resolve them up front. This mirrors `load_indices` in
+        // python/src/dataset.rs, keeping the two listing methods in agreement.
+        let index_type = if let Some(system_type) = infer_system_index_type(example_metadata) {
+            system_type.to_string()
+        } else if details.is_vector() {
+            // Vector indices need to be opened to get the correct type
             let column = field_ids
                 .first()
                 .and_then(|id| dataset.schema().field_by_id(*id))