chore(pre-commit.ci): pre-commit autoupdate

.pre-commit-config.yaml

@@ -10,7 +10,7 @@ repos:
   - id: check-case-conflict
   - id: check-yaml
 - repo: https://github.com/astral-sh/ruff-pre-commit
-  rev: v0.15.12
+  rev: v0.16.8
   hooks:
   - id: ruff-check
   - id: ruff-format
@@ -34,7 +34,7 @@ repos:
     always_run: true
     stages: [pre-push]
 - repo: https://github.com/commitizen-tools/commitizen
-  rev: v4.15.1
+  rev: v4.18.1
   hooks:
   - id: commitizen
     stages: [commit-msg]

docs/roadmaps/phase-3-getitem-glue-audit.md

@@ -244,7 +244,7 @@ pub fn reconstruct_haplotypes_fused<'py>(
 out_data, out_offsets = gvl_rust.reconstruct_haplotypes_fused(
     regions=req.regions,
     geno_offset_idx=req.geno_offset_idx,
-    geno_offsets=self.genotypes.offsets,   # already (2,n) or 1-D; Rust normalizes
+    geno_offsets=self.genotypes.offsets,  # already (2,n) or 1-D; Rust normalizes
     geno_v_idxs=self.genotypes.data,
     v_starts=self.variants.start,
     ilens=self.variants.ilen,

docs/source/dataset.md

@@ -130,7 +130,11 @@ Set `realign_tracks=False` in two cases:
 ds = gvl.get_dummy_dataset()
 
 # Reference-coordinate float tracks alongside haplotypes
-ds_ref_tracks = ds.with_seqs("haplotypes").with_tracks(["read-depth"]).with_settings(realign_tracks=False)
+ds_ref_tracks = (
+    ds.with_seqs("haplotypes")
+    .with_tracks(["read-depth"])
+    .with_settings(realign_tracks=False)
+)
 
 # Interval tracks alongside haplotypes (realign_tracks=False is required)
 ds_itvs = (
@@ -164,7 +168,9 @@ ds = gvl.Dataset.open("ds.gvl", reference="ref.fa", var_fields=["AF"])
 rv = ds.with_seqs("variants")[0, 0]
 rv["AF"]  # per-variant AF values, aligned with rv.alt/.start/.ilen
 
-win = ds.with_seqs("variant-windows", gvl.VarWindowOpt(...)).with_output_format("flat")[0, 0]
+win = ds.with_seqs("variant-windows", gvl.VarWindowOpt(...)).with_output_format("flat")[
+    0, 0
+]
 win.fields["AF"]  # same field, alongside win.fields["start"]/["ilen"]
 ```
 

docs/source/faq.md

@@ -76,7 +76,7 @@ GVL's read path (haplotype reconstruction and track re-alignment) is parallelize
 Environment variables configure a whole process, which means a script's parallelism can't be determined by reading the script — a value in a shell profile, a Dockerfile, or a SLURM template changes how it runs. To state the policy where a reader can see it, set it on the dataset:
 
 ```python
-ds = ds.with_settings(parallel=False)   # True | False | "auto"
+ds = ds.with_settings(parallel=False)  # True | False | "auto"
 ```
 
 - `True` — always hand batches to rayon, whatever their size.

docs/superpowers/REGRESSIONS.md

@@ -148,19 +148,35 @@ cells fit within a ≤256 GB allocation. The OOM-kills observed in the new pipel
 ```python
 import genvarloader as gvl
 from time import perf_counter
+
 # tracks-only: with_seqs(None) -> batch is a single track tensor (no RaggedVariants)
-ds = gvl.Dataset.open(DS_PATH, FASTA).with_seqs(None).with_tracks("read-depth","tracks").with_len(16384)
+ds = (
+    gvl.Dataset.open(DS_PATH, FASTA)
+    .with_seqs(None)
+    .with_tracks("read-depth", "tracks")
+    .with_len(16384)
+)
 for bs in (8, 32):
     dl = ds.to_dataloader(batch_size=bs, shuffle=False)
-    ny=nnuc=0; burn=5; nb=150; t0=perf_counter(); esz=4; done=False
+    ny = nnuc = 0
+    burn = 5
+    nb = 150
+    t0 = perf_counter()
+    esz = 4
+    done = False
     while not done:
         for b in dl:
-            trk = b[1] if isinstance(b,(list,tuple)) else b   # track tensor
-            if ny==burn: t0=perf_counter()
-            if ny>=burn: nnuc+=trk.numel(); esz=trk.element_size()
-            ny+=1
-            if ny>=nb: done=True; break
-    print(bs, nnuc/(perf_counter()-t0)/2**20*esz, "MiB/s")
+            trk = b[1] if isinstance(b, (list, tuple)) else b  # track tensor
+            if ny == burn:
+                t0 = perf_counter()
+            if ny >= burn:
+                nnuc += trk.numel()
+                esz = trk.element_size()
+            ny += 1
+            if ny >= nb:
+                done = True
+                break
+    print(bs, nnuc / (perf_counter() - t0) / 2**20 * esz, "MiB/s")
 ```
 Run with `NUMBA_NUM_THREADS=1` for the single-thread numbers. The 0.6.1 side uses the
 equivalent old API (`gvl.Dataset.open(ds, fasta, return_sequences=False)`; restored in
@@ -171,12 +187,18 @@ the *same* BED + variants + BigWig table to keep regions/samples identical.
 
 ```python
 import numpy as np, numba as nb
+
+
 @nb.njit(parallel=True)
 def f(x):
-    s=0.0
-    for i in nb.prange(x.size): s+=x[i]
+    s = 0.0
+    for i in nb.prange(x.size):
+        s += x[i]
     return s
-f(np.ones(1000)); print(nb.threading_layer())   # 'tbb' if installed, else 'omp'
+
+
+f(np.ones(1000))
+print(nb.threading_layer())  # 'tbb' if installed, else 'omp'
 ```
 0.6.1 pulled `tbb` transitively; 0.24.1 made it optional, so fresh installs report `omp`.
 

docs/superpowers/plans/2026-05-08-get-splice-bed.md

@@ -259,11 +259,13 @@ def get_splice_bed(
     if contigs is not None:
         lf = lf.filter(pl.col("seqname").is_in(contigs))
 
-    lf = lf.filter(pl.col("feature") == "CDS").rename({
-        "seqname": "chrom",
-        "start": "chromStart",
-        "end": "chromEnd",
-    })
+    lf = lf.filter(pl.col("feature") == "CDS").rename(
+        {
+            "seqname": "chrom",
+            "start": "chromStart",
+            "end": "chromEnd",
+        }
+    )
 
     lf = lf.with_columns(
         pl.col("chromStart") - 1,