v1.0.1

The first release since 1.0.0, and mostly a correctness one. The expensive class
of bug here is not a crash: it is a run that returns a matrix that looks
entirely reasonable and is wrong. Most of what follows is that.

📖 Documentation: https://pathogenomics-lab.github.io/distree/ ·
🎓 Tutorial ·
📋 Full changelog

Highlights

Trees that used to fail, or quietly lie

• A leading [&R] / [&U] rooting marker, which IQ-TREE, MrBayes and BEAST all
  write, made a tree unparseable.
• Non-ASCII labels came out as mojibake, so any name with an accent, a Greek
  letter or CJK text stopped matching the samples it came from.
• A truncated tree was accepted, with every dangling internal node silently
  losing its branch length. A file holding several trees produced a matrix for
  the first one without saying so.
• A:1e910 parsed as infinity, giving a matrix of inf with NaN down the
  diagonal and an exit code saying success.
• --midpoint --topology added a hop to every distance crossing the midpoint.
• Negative patristic distances were rounded up to zero, claiming two distinct
  taxa were the same sample.

Faster

Rows are computed in batches rather than one parallel job each, every worker
formats its own output, and fixed-precision formatting is done by hand where the
rounding is unambiguous. An 8,000-tip run went from 2.52 s to 0.13 s on 14
cores; a 20,000-tip matrix from 10.6 s to 1.0 s. Output is byte-identical.

New

• --npy writes a NumPy array of 64-bit floats: exact, half the size of text,
  2.6x faster. With --lower, the condensed vector SciPy reads directly.
• --taxa FILE restricts the matrix to a list of tips, with the distances the
  full tree gives.
• --stats prints a run summary to stderr.
• Gzipped input is read directly, detected by content rather than file name.

Checked against something other than itself

Distances are cross-validated against R's ape (cophenetic.phylo, vcv.phylo)
and phangorn::midpoint over 250 generated trees: agreement to under 1e-9, and
exact for edge counts. The parser and pipeline are fuzzed. 96 tests, up from 31.

Breaking changes

The parser is stricter. All of these were previously accepted:

• A file holding more than one tree is now an error. Split it first.
• A truncated tree is an error rather than being closed at end of input.
• Trailing content after the tree, and unclosed quotes or comments, are errors.
• A label containing whitespace is rejected in --lower mode, because PHYLIP
  readers treat whitespace as the end of the name.
• A branch length that is not finite is rejected.
• --threads 0 and --precision above 30 are rejected.
• Negative patristic distances are reported rather than rounded to zero.

Install

conda install -c bioconda distree

Or download the binary for your platform below. On macOS, clear the quarantine
flag first: xattr -d com.apple.quarantine distree-macos-aarch64.

Platform	Asset
Linux x86-64	distree-linux-x86_64
Linux arm64	distree-linux-aarch64
macOS Apple silicon	distree-macos-aarch64
macOS Intel	distree-macos-x86_64