One-query implementation of search by Codex.

[mooskagh]

search(codex): add minimal PUCT search implementation and register it; hook into build via meson

[borg323]

Quick fix for opponent move display and a few other issues I found:

--- a/src/search/codex/codex.cc
+++ b/src/search/codex/codex.cc
@@ -104,6 +104,7 @@ class CodexSearch : public SearchBase {
     params_ = params;
     // Compute simple time budget.
     ComputeTimeBudget();
+    if (worker_.joinable()) worker_.join();
     worker_ = std::thread(&CodexSearch::SearchLoop, this);
   }
 
@@ -270,6 +271,7 @@ class CodexSearch : public SearchBase {
   // Prepare and send a thinking info update (best-so-far line and score).
   void SendThinkingInfo() {
     if (!root_ || root_->edges.empty()) return;
+    if (responded_bestmove_) return;
     const int64_t elapsed_ms = move_start_time_ ?
         std::chrono::duration_cast<std::chrono::milliseconds>(
             std::chrono::steady_clock::now() - *move_start_time_).count() : 0;
@@ -283,6 +285,11 @@ class CodexSearch : public SearchBase {
           [](const Edge& a, const Edge& b) { return a.n < b.n; });
       if (it == node->edges.end() || it->n == 0) break;
       pv.push_back(it->move);
+      // Temporary compatibility: moves are encoded from current player
+      // perspective, flip before sending when black to move.
+      if (game_state_.CurrentPosition().IsBlackToMove() == (pv.size() & 1)) {
+        pv.back().Flip();
+      }
       node = it->child.get();
       if (pv.size() >= 16) break;
     }
@@ -338,6 +345,7 @@ class CodexSearch : public SearchBase {
     // perspective, flip before sending when black to move.
     if (game_state_.CurrentPosition().IsBlackToMove()) {
       if (!info.bestmove.is_null()) info.bestmove.Flip();
+    } else {
       if (!info.ponder.is_null()) info.ponder.Flip();
     }
     uci_responder_->OutputBestMove(&info);

[Copilot]

Pull Request Overview

This PR introduces a minimal PUCT (Predictor + Upper Confidence bounds applied to Trees) search implementation named "Codex" for the Leela Chess Zero engine. This implementation provides a simple, single-threaded alternative to the main search engine for demonstration and experimentation purposes.

Key Changes:

• Implements a lightweight PUCT tree search using direct Backend API calls with basic node selection, expansion, and backpropagation
• Registers the new search implementation as "codex" via the search factory pattern
• Integrates the implementation into the build system

Reviewed Changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 4 comments.

File	Description
src/search/codex/codex.cc	Complete implementation of CodexSearch class with PUCT algorithm, node/edge structures, and UCI integration
meson.build	Adds codex.cc to the build file list

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Copilot]

Corrected spelling of 'CreateBatchSplitingBackend' to 'CreateBatchSplittingBackend'.

Suggested change

	batchsplit_backend_ = CreateBatchSplitingBackend(backend);
	batchsplit_backend_ = CreateBatchSplittingBackend(backend);

[Copilot]

Division operation inside the selection loop could be optimized by precomputing 1.0f / (1.0f + e.n) or by computing the common factor kCpuct * sqrt_total once before the loop, reducing redundant calculations per child edge.

Suggested change

	int best_idx = -1;
	float best_score = -std::numeric_limits<float>::infinity();
	for (size_t i = 0; i < node->edges.size(); ++i) {
	const Edge& e = node->edges[i];
	const float q = (e.n > 0) ? (e.w / e.n) : 0.0f;
	const float u = kCpuct * e.prior * sqrt_total / (1.0f + e.n);
	const float cpuct_sqrt = kCpuct * sqrt_total;
	int best_idx = -1;
	float best_score = -std::numeric_limits<float>::infinity();
	for (size_t i = 0; i < node->edges.size(); ++i) {
	const Edge& e = node->edges[i];
	const float q = (e.n > 0) ? (e.w / e.n) : 0.0f;
	const float recip = 1.0f / (1.0f + e.n);
	const float u = cpuct_sqrt * e.prior * recip;

[Copilot]

The nodes_visited_ counter is accessed from both the worker thread (incrementing in RunOneSimulation) and potentially the main thread during shutdown/info updates, but is not protected by atomics or mutex. This creates a data race. Consider making it std::atomic<int64_t> or protecting access with the appropriate synchronization mechanism.

Suggested change

	int64_t nodes_visited_ = 0;
	std::atomic<int64_t> nodes_visited_{0};

[Copilot]

The max_depth_ variable is updated in RunOneSimulation (line 251) on the worker thread and read in SendThinkingInfo without synchronization. This is a data race. Consider using std::atomic<int> for thread-safe access.

Suggested change

	int max_depth_ = 0;
	std::atomic<int> max_depth_{0};