Skip to content

dead_code_report_current.md

Latest commit

 

History

History
64 lines (44 loc) · 3.25 KB

dead_code_report_current.md

File metadata and controls

64 lines (44 loc) · 3.25 KB

Dead-code report — current status

Status: 2026-05-01, after cleanup on the current working tree.

This supersedes the earlier dead_code_report.md from the Telegram temp directory. The original report was useful, but several findings are now either fixed, removed, or reclassified as intentional engine surface.

Resolved

The following zero-caller or orphaned paths from the old report have been removed:

Old finding Resolution
ray_poll_rx_extend / ray_poll_send Removed from src/core/poll.c and src/core/poll.h. IPC continues to use its direct send path.
ray_part_load Removed from src/store/part.c and src/store/part.h. The supported path is ray_read_parted.
Embedding DAG ops: exec_cosine_sim, exec_euclidean_dist, exec_knn, exec_hnsw_knn Removed with their graph builders, exec dispatch, stale opcodes, dump labels, and unused extension-union fields. Live Rayfall embedding builtins remain.
OP_TIL DAG path Removed the graph builder, exec dispatch, and optimizer fold. The live Rayfall (til ...) builtin remains.
OP_PROD grouped aggregation Fixed accumulation, merge, and emit paths. Added rfl regression coverage for scalar, DA, and HT paths.

Verification:

rg -n 'ray_part_load|ray_poll_send|ray_poll_rx_extend|exec_cosine_sim|exec_euclidean_dist|exec_knn\(|exec_hnsw_knn|ray_cosine_sim|ray_euclidean_dist|ray_knn\(|ray_hnsw_knn|OP_TIL|OP_COSINE_SIM|OP_EUCLIDEAN_DIST|OP_HNSW_KNN|OP_KNN\b' src test

No matches remain for the removed symbols.

make test

Result: 991 of 992 passed (1 skipped, 0 failed).

Current diff shape:

15 files changed, 98 insertions(+), 727 deletions(-)

Reclassified As Live / Intentional

These old report items should not be deleted.

Area Current status
OP_WINDOW / ray_window_op Live C-level operator with tests in test/test_audit.c and test/test_exec.c. Still not broadly exposed as a Rayfall verb, but not dead.
Graph algorithms (OP_PAGERANK, OP_LOUVAIN, OP_DIJKSTRA, OP_DFS, OP_TOPSORT, centrality ops, MST, random walk, A*, k-shortest, expand, var-expand, shortest path, WCO join) Live engine/C-API surface with coverage in test/test_csr.c and related tests. Public Rayfall exposure is a product/API decision, not dead-code cleanup.
MAPCOMMON / parted execution paths Live storage/execution surface. Covered by test/test_store.c, test/test_opt.c, test/test_link.c, and rfl system tests.
Factorized expand/group paths Live C-level graph pipeline with tests in test/test_csr.c.
REPL-only syscmd handlers Live in interactive mode. Coverage should come from REPL fixtures rather than deletion.

Remaining Cleanup Candidates

No old high-confidence zero-caller removal candidate remains from the report.

The next useful cleanup pass should target one of these:

  1. Add or improve REPL fixture coverage for src/lang/syscmd.c.
  2. Decide which C-level graph algorithms should get Rayfall verbs before first users.
  3. Decide whether OP_WINDOW should become a Rayfall-facing feature now or stay C-level.
  4. Continue a broader zero-caller audit outside the original report scope, but treat C-API/test-only graph operators as intentional unless there is a design decision to remove them.