src/obidebruinj/src/debruijn.rs

@@ -1,15 +1,15 @@
 //use ahash::RandomState;
+use crossbeam_channel;
 use hashbrown::HashMap;
 use obikseq::k;
 use obikseq::{CanonicalKmer, Sequence, Unitig};
 use rayon::iter::{IntoParallelRefIterator, ParallelIterator};
 use std::cell::RefCell;
 use std::fmt;
-use crossbeam_channel;
 use std::sync::atomic::{AtomicU8, Ordering};
-use xxhash_rust::xxh3::Xxh3Builder;
 use std::time::Instant;
 use tracing::{debug, info};
+use xxhash_rust::xxh3::Xxh3Builder;
 
 // ── Types ─────────────────────────────────────────────────────────────────────
 
@@ -99,7 +99,6 @@ impl Node {
         (self.0 >> 5) & 0b11
     }
 
-
     /// Marks the node as visited.
     #[inline]
     pub fn set_visited(&mut self) {
@@ -180,7 +179,11 @@ impl WalkState {
     }
 
     pub fn reachable(&self, graph: &GraphDeBruijn) -> bool {
-        WalkState { kmer: self.kmer, node: self.node, direct: !self.direct }
+        WalkState {
+            kmer: self.kmer,
+            node: self.node,
+            direct: !self.direct,
+        }
         .leavable(graph)
     }
 
@@ -197,8 +200,19 @@ impl WalkState {
             if next_node.is_visited() {
                 return None;
             }
-            let reachable = if dnext { next_node.can_extend_left() } else { next_node.can_extend_right() };
+            let reachable = if dnext {
-            reachable.then_some((WalkState { kmer: cnext, node: next_node, direct: dnext }, nuc))
+                next_node.can_extend_left()
+            } else {
+                next_node.can_extend_right()
+            };
+            reachable.then_some((
+                WalkState {
+                    kmer: cnext,
+                    node: next_node,
+                    direct: dnext,
+                },
+                nuc,
+            ))
         } else {
             if !self.node.can_extend_left() {
                 return None;
@@ -211,8 +225,19 @@ impl WalkState {
             if next_node.is_visited() {
                 return None;
             }
-            let reachable = if dnext { next_node.can_extend_right() } else { next_node.can_extend_left() };
+            let reachable = if dnext {
-            reachable.then_some((WalkState { kmer: cnext, node: next_node, direct: dnext }, 3 - nuc))
+                next_node.can_extend_right()
+            } else {
+                next_node.can_extend_left()
+            };
+            reachable.then_some((
+                WalkState {
+                    kmer: cnext,
+                    node: next_node,
+                    direct: dnext,
+                },
+                3 - nuc,
+            ))
         }
     }
 }
@@ -275,7 +300,11 @@ impl GraphDeBruijn {
             node.set_left(lc, ln);
             atomic.store(node.0, Ordering::Relaxed);
         });
-        debug!("[compute_degrees] pass 1 (degrees): {:?} — {} nodes", t1.elapsed(), self.nodes.len());
+        debug!(
+            "[compute_degrees] pass 1 (degrees): {:?} — {} nodes",
+            t1.elapsed(),
+            self.nodes.len()
+        );
 
         // Pass 2: mark start nodes
 
@@ -290,7 +319,11 @@ impl GraphDeBruijn {
                 atomic.store(node.0, Ordering::Relaxed);
             }
         });
-        debug!("[compute_degrees] pass 2 (starts):  {:?} — {} nodes", t2.elapsed(), self.nodes.len());
+        debug!(
+            "[compute_degrees] pass 2 (starts):  {:?} — {} nodes",
+            t2.elapsed(),
+            self.nodes.len()
+        );
     }
 
     pub fn is_visited(&self, kmer: &CanonicalKmer) -> Option<bool> {
@@ -328,14 +361,28 @@ impl GraphDeBruijn {
     }
 
     fn unitig_nucleotides(&self, kmer: CanonicalKmer, k: usize) -> Option<UnitigNucIter<'_>> {
-        let old = self.nodes.get(&kmer)?.fetch_or(IS_VISITED_MASK, Ordering::AcqRel);
+        let old = self
-        if old & IS_VISITED_MASK != 0 { return None; }
+            .nodes
+            .get(&kmer)?
+            .fetch_or(IS_VISITED_MASK, Ordering::AcqRel);
+        if old & IS_VISITED_MASK != 0 {
+            return None;
+        }
         let start = WalkState::new(kmer, Node(old), true);
         let next_step = start.walk(self).and_then(|(next_state, nuc)| {
-            let ext_old = self.nodes.get(&next_state.kmer)?.fetch_or(IS_VISITED_MASK, Ordering::AcqRel);
+            let ext_old = self
+                .nodes
+                .get(&next_state.kmer)?
+                .fetch_or(IS_VISITED_MASK, Ordering::AcqRel);
             (ext_old & IS_VISITED_MASK == 0).then_some((next_state, nuc))
         });
-        Some(UnitigNucIter { graph: self, start: kmer, pos: 0, k, next_step })
+        Some(UnitigNucIter {
+            graph: self,
+            start: kmer,
+            pos: 0,
+            k,
+            next_step,
+        })
     }
 
     pub fn for_each_unitig(&self, f: impl Fn(UnitigNucIter<'_>) + Sync) {
@@ -352,7 +399,9 @@ impl GraphDeBruijn {
             self.nodes
                 .par_iter()
                 .filter_map(|(&kmer, atomic)| {
-                    Node(atomic.load(Ordering::Acquire)).is_start().then_some(kmer)
+                    Node(atomic.load(Ordering::Acquire))
+                        .is_start()
+                        .then_some(kmer)
                 })
                 .for_each(|kmer| {
                     if let Some(iter) = self.unitig_nucleotides(kmer, k) {
@@ -403,7 +452,7 @@ impl GraphDeBruijn {
             }
         }
 
-        info!(
+        debug!(
             chains = n_chains.load(Ordering::Relaxed),
             phase2 = n2.load(Ordering::Relaxed),
             total = n_chains.load(Ordering::Relaxed) + n2.load(Ordering::Relaxed),
@@ -508,7 +557,11 @@ impl Iterator for UnitigNucIter<'_> {
             Some(nuc)
         } else if let Some((state, nuc)) = self.next_step.take() {
             self.next_step = state.walk(self.graph).and_then(|(next_state, next_nuc)| {
-                let old = self.graph.nodes.get(&next_state.kmer)?.fetch_or(IS_VISITED_MASK, Ordering::AcqRel);
+                let old = self
+                    .graph
+                    .nodes
+                    .get(&next_state.kmer)?
+                    .fetch_or(IS_VISITED_MASK, Ordering::AcqRel);
                 (old & IS_VISITED_MASK == 0).then_some((next_state, next_nuc))
             });
             Some(nuc)
@@ -539,10 +592,16 @@ fn count_neighbors(
             }
             nuc = i as u8;
             count += 1;
-            if count >= 2 { return (2, None); }
+            if count >= 2 {
+                return (2, None);
             }
         }
-    if count == 1 { (1, Some(nuc)) } else { (0, None) }
+    }
+    if count == 1 {
+        (1, Some(nuc))
+    } else {
+        (0, None)
+    }
 }
 
 // ── tests ─────────────────────────────────────────────────────────────────────

src/obikindex/src/merge.rs

@@ -65,39 +65,65 @@ impl KmerIndex {
         }
 
         // ── Log source characteristics and choose base ────────────────────────
-        let mode_str = if mode == MergeMode::Presence { "presence" } else { "count" };
+        let mode_str = if mode == MergeMode::Presence {
+            "presence"
+        } else {
+            "count"
+        };
         info!(
             "merge: {} source(s), smer-size={}, mode={}",
-            sources.len(), sources[0].kmer_size(), mode_str,
+            sources.len(),
+            sources[0].kmer_size(),
+            mode_str,
         );
         for (i, src) in sources.iter().enumerate() {
-            let genome_str = if src.meta.genomes.len() == 1 { "mono-genome".to_string() }
+            let genome_str = if src.meta.genomes.len() == 1 {
-                             else { format!("{} genomes", src.meta.genomes.len()) };
+                "mono-genome".to_string()
-            let trivial_str = if is_trivial(src, mode) { " [trivial: no data approximation]" } else { "" };
+            } else {
+                format!("{} genomes", src.meta.genomes.len())
+            };
+            let trivial_str = if is_trivial(src, mode) {
+                " [trivial: no data approximation]"
+            } else {
+                ""
+            };
             info!(
                 "  [{}] {} — {}, {}, {}{}",
-                i, src.root_path.display(),
+                i,
+                src.root_path.display(),
                 format_evidence(&src.meta.config.evidence),
-                genome_str, mode_str, trivial_str,
+                genome_str,
+                mode_str,
+                trivial_str,
             );
         }
 
         let base_idx = choose_base(sources, mode);
         let needs_approx = sources.iter().any(|src| {
             !is_trivial(src, mode)
-                && matches!(src.meta.config.evidence, IndexMode::Approx { .. } | IndexMode::Hybrid { .. })
+                && matches!(
+                    src.meta.config.evidence,
+                    IndexMode::Approx { .. } | IndexMode::Hybrid { .. }
+                )
         });
         info!(
             "output evidence: {} ({}base: [{}] {})",
             format_evidence(&sources[base_idx].meta.config.evidence),
-            if needs_approx { "forced approx — " } else { "" },
+            if needs_approx {
-            base_idx, sources[base_idx].root_path.display(),
+                "forced approx — "
+            } else {
+                ""
+            },
+            base_idx,
+            sources[base_idx].root_path.display(),
         );
 
         let mut ordered: Vec<&KmerIndex> = Vec::with_capacity(sources.len());
         ordered.push(sources[base_idx]);
         for (i, &src) in sources.iter().enumerate() {
-            if i != base_idx { ordered.push(src); }
+            if i != base_idx {
+                ordered.push(src);
+            }
         }
         let sources: &[&KmerIndex] = &ordered;
         let evidence = sources[0].meta.config.evidence.clone();
@@ -151,7 +177,8 @@ impl KmerIndex {
             fs::remove_dir_all(&spectrums_dir)?;
         }
         for (src, new_labels) in sources.iter().zip(&source_labels) {
-            let old_labels: Vec<String> = src.meta.genomes.iter().map(|g| g.label.clone()).collect();
+            let old_labels: Vec<String> =
+                src.meta.genomes.iter().map(|g| g.label.clone()).collect();
             copy_spectrums(&src.root_path, output, &old_labels, new_labels)?;
         }
         pb.finish_and_clear();
@@ -184,9 +211,12 @@ impl KmerIndex {
 
             // Per-partition unitig byte sizes across remaining sources (stat() only)
             let partition_sizes: Vec<u64> = (0..n_partitions)
-                .map(|i| remaining_sources.iter()
+                .map(|i| {
+                    remaining_sources
+                        .iter()
                         .map(|s| partition_unitig_bytes(s, i))
-                    .sum())
+                        .sum()
+                })
                 .collect();
 
             // LFD sort: largest partition first
@@ -201,7 +231,8 @@ impl KmerIndex {
             // IDs; each reports (id, g_len, duration) on a result channel.
             const SPAWN_THRESHOLD: f64 = 0.95; // spawn when >5% capacity idle
             let n_cores = std::thread::available_parallelism()
-                .map(|n| n.get()).unwrap_or(1);
+                .map(|n| n.get())
+                .unwrap_or(1);
             let max_workers = (n_cores / 2).max(1);
             let _ = budget_fraction; // kept in signature for CLI compatibility
 
@@ -220,6 +251,9 @@ impl KmerIndex {
             let mut part_stats: Vec<PartStat> = Vec::with_capacity(n_partitions);
             let mut n_workers = 0usize;
             let mut cpu_sample = CpuSample::now();
+            // Efficiency measured just before each spawn, used to assess
+            // whether the previous worker delivered its expected marginal gain.
+            let mut efficiency_at_last_spawn = 0.0f64;
 
             // Shadow as references so closures can capture them by copy.
             let srcs = &srcs;
@@ -237,7 +271,12 @@ impl KmerIndex {
                             for i in &prx {
                                 let t = Instant::now();
                                 let r = dst_partition.merge_partition(
-                                    i, srcs, mode, n_dst_genomes, block_bits, evidence,
+                                    i,
+                                    srcs,
+                                    mode,
+                                    n_dst_genomes,
+                                    block_bits,
+                                    evidence,
                                 );
                                 rtx.send((i, r, t.elapsed())).ok();
                             }
@@ -252,26 +291,51 @@ impl KmerIndex {
 
                 let mut completed = 0usize;
                 while completed < n_partitions {
-                    let (i, r, dur) = result_rx.recv()
+                    let (i, r, dur) = result_rx.recv().map_err(|_| {
-                        .map_err(|_| OKIError::Io(io::Error::new(
+                        OKIError::Io(io::Error::new(
-                            io::ErrorKind::UnexpectedEof, "worker channel closed")))?;
+                            io::ErrorKind::UnexpectedEof,
+                            "worker channel closed",
+                        ))
+                    })?;
                     let g_len = r.map_err(OKIError::Partition)?;
                     pb.inc(1);
-                    debug!("partition {i}: done in {:.1}s — {} new kmers",
+                    debug!(
-                        dur.as_secs_f64(), g_len);
+                        "partition {i}: done in {:.1}s — {} new kmers",
+                        dur.as_secs_f64(),
+                        g_len
+                    );
                     part_stats.push(PartStat {
-                        id: i, unitig_bytes: partition_sizes[i], g_len,
+                        id: i,
+                        unitig_bytes: partition_sizes[i],
+                        g_len,
                     });
                     completed += 1;
 
                     if n_workers < max_workers && completed < n_partitions {
                         let eff = cpu_sample.cpu_efficiency(n_cores);
-                        if eff < SPAWN_THRESHOLD {
+                        // For the first spawn use SPAWN_THRESHOLD.
+                        // For subsequent spawns: the previous worker should
+                        // have raised efficiency by at least a quarter of the expected
+                        // marginal gain (1/n_workers). If not, adding another
+                        // worker won't help.
+                        let should_spawn = if n_workers == 1 {
+                            eff < SPAWN_THRESHOLD
+                        } else {
+                            let gain = eff - efficiency_at_last_spawn;
+                            let expected = 1.0 / n_workers as f64;
+                            gain >= expected * 0.25
+                        };
+                        if should_spawn {
+                            debug!(
+                                "activated worker {} — efficiency {:.0}%, gain vs prev {:.0}%",
+                                n_workers + 1,
+                                eff * 100.0,
+                                (eff - efficiency_at_last_spawn) * 100.0,
+                            );
+                            efficiency_at_last_spawn = eff;
                             activate_tx.send(()).ok();
                             n_workers += 1;
                             cpu_sample = CpuSample::now();
-                            debug!("activated worker {n_workers} — efficiency {:.0}%",
-                                eff * 100.0);
                         }
                     }
                 }
@@ -319,9 +383,7 @@ fn print_merge_partition_report(stats: &[PartStat], n_workers: usize, max_worker
         "  {} partition(s) processed, {} total new kmers",
         non_empty, total_new,
     );
-    info!(
+    info!("  workers spawned: {n_workers} / {max_workers} (max)",);
-        "  workers spawned: {n_workers} / {max_workers} (max)",
-    );
 
     // Top 8 partitions by new-kmer count
     let mut by_new: Vec<&PartStat> = stats.iter().filter(|s| s.g_len > 0).collect();
@@ -343,10 +405,15 @@ fn print_merge_partition_report(stats: &[PartStat], n_workers: usize, max_worker
 // ── helpers ───────────────────────────────────────────────────────────────────
 
 fn fmt_bytes(b: u64) -> String {
-    if b >= 1 << 30      { format!("{:.1} GB", b as f64 / (1u64 << 30) as f64) }
+    if b >= 1 << 30 {
-    else if b >= 1 << 20 { format!("{:.1} MB", b as f64 / (1u64 << 20) as f64) }
+        format!("{:.1} GB", b as f64 / (1u64 << 30) as f64)
-    else if b >= 1 << 10 { format!("{:.1} KB", b as f64 / (1u64 << 10) as f64) }
+    } else if b >= 1 << 20 {
-    else                 { format!("{b} B") }
+        format!("{:.1} MB", b as f64 / (1u64 << 20) as f64)
+    } else if b >= 1 << 10 {
+        format!("{:.1} KB", b as f64 / (1u64 << 10) as f64)
+    } else {
+        format!("{b} B")
+    }
 }
 
 /// Sum of all unitigs.bin sizes across all layers of partition `i` in `src`.
@@ -354,8 +421,12 @@ fn partition_unitig_bytes(src: &KmerIndex, i: usize) -> u64 {
     let mut total = 0u64;
     for l in 0.. {
         let p = src.layer_unitigs_path(i, l);
-        if !p.exists() { break; }
+        if !p.exists() {
-        if let Ok(m) = std::fs::metadata(&p) { total += m.len(); }
+            break;
+        }
+        if let Ok(m) = std::fs::metadata(&p) {
+            total += m.len();
+        }
     }
     total
 }
@@ -382,7 +453,10 @@ fn compute_labels(
             };
             *count += 1;
             labels.push(new_label.clone());
-            all_genomes.push(GenomeInfo { label: new_label, meta: genome.meta.clone() });
+            all_genomes.push(GenomeInfo {
+                label: new_label,
+                meta: genome.meta.clone(),
+            });
         }
         source_labels.push(labels);
     }
@@ -443,13 +517,21 @@ fn index_unitig_size(src: &KmerIndex) -> u64 {
 fn choose_base(sources: &[&KmerIndex], mode: MergeMode) -> usize {
     let needs_approx = sources.iter().any(|src| {
         !is_trivial(src, mode)
-            && matches!(src.meta.config.evidence, IndexMode::Approx { .. } | IndexMode::Hybrid { .. })
+            && matches!(
+                src.meta.config.evidence,
+                IndexMode::Approx { .. } | IndexMode::Hybrid { .. }
+            )
     });
 
-    sources.iter().enumerate()
+    sources
+        .iter()
+        .enumerate()
         .filter(|(_, src)| {
             !needs_approx
-                || matches!(src.meta.config.evidence, IndexMode::Approx { .. } | IndexMode::Hybrid { .. })
+                || matches!(
+                    src.meta.config.evidence,
+                    IndexMode::Approx { .. } | IndexMode::Hybrid { .. }
+                )
         })
         .max_by_key(|(_, src)| index_unitig_size(src))
         .map(|(i, _)| i)