v1.6.0

A broad feature release on top of 1.5.0. Highlights: bigBed / bigWig
and 2bit support via a vendored libBigWig and a hand-rolled 2bit
reader; full TUI exploration suite (column stats, sort, show/hide,
live filter, copy cell); generic Parquet range queries with column
auto-detection; markdown output and LociSSD validation; coordinate-
convention selector with UCSC and NCBI labels; color themes; LTO +
ccache static builds; and ARM64 release artifacts.

Changed

• Smarter inline truncation for list / map cells. The
  truncator now walks every top-level comma and picks the largest
  leading-element prefix that fits the column. Previous behavior
  always dropped to [first, …] after the first comma, even when
  [first, second] would have fit. Most visible on VCF INFO
  expansions, BAM tags, and other multi-element list cells common
  in real data. Falls back to char-truncation only when even one
  element + ellipsis doesn't fit.

Added

• TUI copy cell (y key) — copies the top-left visible cell to
  the system clipboard via OSC52. Works through SSH and tmux 3.3+
  with no xclip / pbcopy helper required (modern terminal
  emulators intercept the escape directly). Status bar flashes
  copied: <preview> until the next key; auto-clears.

• UCSC 2bit (.2bit) — sequence container used for genome
  references (hg38.2bit, mm10.2bit). vv exposes the sequence index
  rather than decoded bases (chromosome-scale strings are too large
  to materialise): name, length, n_blocks (unknown-base runs),
  mask_blocks (soft-masked-region runs). Endianness is detected
  from the signature; long-offset 64-bit 2bit is rejected with a
  clear message (use twoBitToFa for that variant).

• --theme <name> — color palette selector. Built-in themes:
  default, dark, light, solarized-dark, solarized-light
  (solarized accepted as a synonym for solarized-dark). Affects
  both the non-interactive ASCII table (ANSI escapes) and the
  ncurses TUI. Each theme provides a 256-color palette and a
  16-color fallback for older terminals; the light theme also
  suppresses zebra stripes so the table reads cleanly on a bright
  background. Unknown theme names produce a clear error listing the
  available choices.

• TUI live filter (& key) — long-requested. Opens a
  &<expression> input bar; expression grammar matches the
  --filter CLI flag. Rows not matching the predicate are hidden
  from the display; status bar shows filter:<expr> N/M (rows
  kept / total). Composes with sort (s): one full-file pass
  rebuilds the combined view via the existing source_row(display)
  indirection. Empty input or Esc (in normal mode) clears.

  Internals: a new rebuild_display_order() orchestrates both
  sort and filter into the single sort_order_ permutation. When
  both are active, rows that pass the filter are collected with
  their sort keys in one pass and sorted at the end.

• --decode-threads N — separately size Arrow's CPU thread pool
  (Parquet column decode and CSV / TSV parallel parsing) without
  touching htslib's thread count. Defaults to --threads; useful
  when cold reads bottleneck on column decompression. Bounded at
  2 × hardware_concurrency().

Changed

• Release pipeline: now builds both x86_64 and aarch64
  (ARM64) static Linux binaries on every tag. The release workflow
  matrices over ubuntu-latest and the new GitHub-hosted
  ubuntu-22.04-arm runner — native builds on each architecture,
  no QEMU emulation. Tarballs are named
  vv-<ver>-linux-<arch>.tar.gz; a merged SHA256SUMS covers both.
• AlmaLinux 8 Docker build: dropped the -march=x86-64
  baseline flag from the build-app CMake invocation so the same
  Dockerfile builds on aarch64. (-march=x86-64 is the default
  target on x86_64 gcc anyway, so the change is a no-op there.)
• AlmaLinux 8 Docker build: rebuild every static dep + vv itself
  with -flto=auto. gcc-ar / gcc-ranlib preserve LTO IR in archives
  so the final link can drop unused code across library boundaries
  (not just within each .a). Significant text-segment shrink vs
  the previous non-LTO 15 MB static binary.
• AlmaLinux 8 Docker build: install ccache (via EPEL) and put
  its compiler-symlink dir ahead of gcc-toolset-12 on PATH. A
  BuildKit cache mount (id=vv-ccache) persists the compile cache
  across docker builds. Requires Docker BuildKit + buildx.

Added

• TUI interactive-exploration batch — three new keybindings turn
  the ncurses viewer into a real exploration tool:
  • S opens a per-column stats popup (count, nulls, min, max,
    mean, distinct count) for the leftmost visible column. Full-file
    scan with progress in the status line.
  • s sorts by the leftmost visible column. Repeated s on the
    same column toggles ascending / descending; s on a different
    column re-sorts ascending; u clears the sort. Numeric types
    sort by raw value, others lexicographically; nulls last. Search
    and the detail pane follow the sorted display order via a single
    source_row(display) indirection that routes all cache lookups
    through the sort permutation when active.
  • c opens a column show/hide picker overlay; Space toggles
    the column under the cursor, hidden columns drop from the
    layout, and the status bar shows a hidden:N indicator.
• --tail N — show the last N rows instead of the first N.
  Mirrors -n but operates on the tail. Reuses the
  MemoryTableSource adapter: every chunk is read (through any
  active --filter) and the last N rows are sliced off, so the
  result renders identically through every view / export mode
  (table, TUI, TSV, CSV, JSON, Markdown, Parquet).
• --coords UCSC|NCBI — pick the coordinate convention for
  -r. UCSC (default; aliases 0-based, bed) means 0-based
  half-open as in BED, bigBed, bigWig, BAM and LociSSD — the
  convention introduced by Jim Kent's UCSC Genome Browser source
  tree in 2000. NCBI (aliases GenBank, 1-based, tabix)
  means 1-based inclusive as in GenBank (1982), VCF, GFF, and the
  samtools / bcftools / tabix command lines. vv -r chr1:101-200 --coords NCBI file matches the same rows as the UCSC default
  vv -r chr1:100-200 file. Conversion happens once in
  apply_region_modifiers; downstream sources see normalized
  UCSC. --regions-file is always BED (UCSC) per the spec.
• --parquet - — write Parquet to stdout. Parquet's
  footer-at-end requires seekable writes, so the data is spooled
  to an mkstemps temp file under /tmp, then streamed to stdout
  and unlinked. Bit-identical to writing to a file path. Enables
  vv ... --parquet - | tool pipelines.
• Generic Parquet range queries — -r chr1:1000-2000 file.parquet
  now works on any Parquet file with chrom/start/end columns, not just
  LociSSD. Chromosome / Start / End column names are auto-detected
  (common variants: Chromosome/Chrom/Chr, Start/POS/
  chromStart, End/Stop/chromEnd). When auto-detection isn't
  clear (or your columns have unusual names) use
  --region-cols Chr,Start,End. Row-group pruning uses Parquet
  ByteArray statistics on chrom plus Int statistics on Start/End;
  per-row filtering inside surviving row groups is correct for
  unsorted files too, just less efficient. Dictionary-encoded chrom
  columns are handled transparently.
• --validate — LociSSD invariants checker. Verifies the
  lociSSD_manifest footer parses, the required columns
  (Chromosome / Start / End / MaxEndSoFar) have the right types,
  the manifest's per-chromosome row ranges cover every row
  contiguously, rows are sorted by (Start, End) within each
  chromosome, MaxEndSoFar[i] matches max(End[chrom_first..i]),
  and every row's Chromosome label agrees with the manifest's
  window. Prints PASS / FAIL per check, caps repeat violations
  at 5 lines per category, exits non-zero on any failure.
• --md / --markdown — write a GitHub-flavored markdown table
  to stdout. Useful for pasting query results into issues, READMEs,
  and reports. Cells escape | as \| and turn embedded newlines
  into <br>. Honours --select, --filter, -n, --no-header,
  -r, --sample. Numbers keep their _ thousands separators
  so the rendered table stays readable.
• bigBed / bigWig support via a vendored copy of
  libBigWig (0.4.8, MIT-licensed,
  compiled with -DNOCURL so only zlib is required). Source lives
  under vendored/libBigWig/; CMake compiles it as a private static
  library (libbigwig.a) and links it into vv. The static AlmaLinux 8
  Docker build picks it up automatically — no new external dependency
  for any distribution channel.
• bigBed: the embedded autoSql definition is parsed into typed
  Arrow columns (signalValue, pValue, qValue, …). A small
  ~150-LOC hand-rolled parser handles primitives, char[N] fixed-
  width strings, enum{…} / set{…} (as strings), and arrays
  (int[N] / int[fieldName] → Arrow list<int>).
• bigWig: rendered as four columns (chrom, start, end, value).
• Range queries (-r chr1:78-99) use libBigWig's overlap APIs
  (bbGetOverlappingEntries / bwGetOverlappingIntervals); no
  external index needed.
• User manual at docs/USAGE.md, with worked examples for every
  flag, every output mode, the LociSSD / tabix / BCF range-query
  flows, and a "by-workflow" cookbook section. docs/build_docs.sh
  renders it to self-contained HTML and PDF via pandoc (with a
  texlive → chromium-via-HTML fallback).
• BCF range queries — -r chr1:100-200 file.bcf now works, using
  bcf_itr_querys over an existing .csi / .tbi index. Missing
  index produces a clear hint pointing at bcftools index.
• --regions-file <BED> — read additional windows from a BED file's
  first three columns. Combines with explicit -r; both are taken.
• --slop N — pad every window by N bp on each side (start clamped
  at 0). Equivalent to bedtools slop inline.
• --stats — Parquet metadata dump (row groups, codecs, per-column
  compressed / uncompressed sizes, null counts) without reading any
  data. Falls back to --schema for non-Parquet sources.
• --unique COL[,COL,...] — distinct value counts per column.
  Top-50 values per column with overflow indicator. Honours --filter.
• --sample N — reservoir-sample N rows uniformly instead of
  head-N. Honours --filter (filter first, then sample). Combines
  with every view / export mode through a new in-memory
  MemoryTableSource adapter that wraps the sampled Arrow Table and
  routes it through the normal rendering pipeline.
• --schema — print the schema + file metadata footer and exit.
  Cheap, useful for inherited files.
• --describe — per-column statistics: count, nulls, min, max,
  mean (numeric), distinct count (string, capped at 16). Respects
  --select and --filter.
• Column projection by name — --select Chromosome,Start,Score.
  Unknown names produce a clear error. Works across the ASCII table,
  vertical-head, TSV, CSV, JSON, and Parquet output paths. Display
  modes still hide format-derived columns (e.g. LociSSD's
  MaxEndSoFar); export modes preserve the user's explicit list.
• --filter <expr> — value-predicate filter: <col> <op> <value>
  joined by AND / OR. Ops: == != < <= > >=. Values: integer,
  float, single- or double-quoted string. Evaluated per row.
• JSON / NDJSON output — --json writes a JSON array, --ndjson
  writes one JSON object per row. Streams chunk-by-chunk.
• LociSSD range queries: -r chr1:78-99 file.lociss (also
  --window). Pruning uses the embedded manifest to locate the
  chromosome's row range, Parquet column statistics on Start and
  MaxEndSoFar to skip row groups outside the window, and a per-row
  predicate (Start < end AND End > start) inside surviving row
  groups. Open-ended forms (chr1:, chr1:78-, chr1:-99) and
  comma-separated multi-windows supported. Plain Parquet without a
  LociSSD manifest is rejected with a hint.
• TUI: help overlay (H or F1) — centred panel listing every
  keybinding. Any key dismisses it.
• TUI: frozen first column (z toggle) — pin column 0 at the left
  edge while scrolling sideways. Useful for wide tables where the key
  column (CHROM, Name, ...) should stay visible.
• TUI: mouse-wheel scrolling — scroll three rows per wheel notch.
• LociSSD support — Parquet files carrying a lociSSD_manifest key
  in their footer (see FORMAT_SPEC.md) are detected
  automatically. The derived MaxEndSoFar technical column is hidden
  from the ASCII table, the vertical-head view, and the interactive TUI;
  it remains intact in --tsv, --csv, and --parquet output so the
  data round-trips losslessly. New virtual hidden_for_display() hook
  on TabularSource provides the mechanism.