v2.3.2

salmon 2.3.2

A performance-and-polish patch on 2.3.1. No index rebuild is required and
quant.sf is unchanged. Mapping output is byte-identical to 2.3.1 (verified on
real data), and the --deterministic guarantee — byte-identical results across
thread counts — is preserved and unchanged from 2.3.1.

Faster mapping

• Length-dispatched radix sort for uni-MEM grouping (#1044). The per-read
  grouping of uni-MEMs by (transcript, orientation) now uses a radix sort sized
  to the key range instead of a comparison sort, shaving ~2% off mapping time on
  typical bulk runs. The change is purely a hot-path optimization: results are
  output-neutral (byte-identical quant.sf versus 2.3.1 at single thread, and
  byte-identical --deterministic output across thread counts).

Richer meta_info.json and input diagnostics

• More run metadata in aux_info/meta_info.json (#1044). The file gains
  several additive fields — detected_library_type, num_orphan,
  em_converged, num_em_iterations, range_factorization_bins,
  pos_bias_correct, peak_rss_kb, total_time_seconds — and a structured
  diagnostics array. All existing keys are unchanged, so downstream parsers
  (e.g. tximport, pyroe) keep working.
• Always-on library-type-mismatch detection (#1044). salmon now always
  evaluates whether the observed fragment orientations match the requested
  library type and surfaces it via detected_library_type / diagnostics. It is
  a pure observer — it does not change quantification.
• The richer metadata and zero-cost bad-input diagnostics now also cover the
  alignment (-a) and --rad input paths.

Configurable index-build scratch (--sshashTmpDir, --ramLimit)

• salmon index --sshashTmpDir <DIR> (#1048) controls where the SSHash
  minimizer sort writes its on-disk scratch. It now defaults to a sshash_tmp
  subdirectory of --tmpdir (or the index output directory when --tmpdir is
  unset) and is cleaned up after the build — fixing the previous behavior of
  leaving an empty sshash_tmp/ directory in the current working directory. Use
  the flag to place the sort scratch on a separate or faster disk.
• salmon index --ramLimit <GiB> (#1048) caps the RAM the external minimizer
  sort may use (default 8); a smaller value uses less memory but spills to disk
  sooner. This is a cohesive replacement for the fix in #1045 (thanks
  @BenjaminDEMAILLE, whose report motivated
  the change).

Other changes

• Clarified salmon quant --threads help text to note the auxiliary I/O
  thread (#1047, #993) — thanks @BenjaminDEMAILLE.
• Internal: salmon is now available in homebrew-core (brew install salmon),
  so the separate brewsci auto-bump workflow was removed.

Upgrading

Drop-in from 2.3.1: no index rebuild, and quantification results are unchanged.
The new --sshashTmpDir / --ramLimit options only affect salmon index
resource placement, not index content.

---

Install salmon-cli 2.3.2

Install prebuilt binaries via shell script

curl --proto '=https' --tlsv1.2 -LsSf https://github.com/COMBINE-lab/salmon/releases/download/v2.3.2/salmon-cli-installer.sh | sh

Download salmon-cli 2.3.2

File	Platform	Checksum
salmon-cli-aarch64-apple-darwin.tar.xz	Apple Silicon macOS	checksum
salmon-cli-x86_64-apple-darwin.tar.xz	Intel macOS	checksum
salmon-cli-aarch64-unknown-linux-gnu.tar.xz	ARM64 Linux	checksum
salmon-cli-x86_64-unknown-linux-gnu.tar.xz	x64 Linux	checksum