Zig drop-in replacement for bcrham and ig-sw (zero C++ deps)

[matsen]

Summary

Wires in matsengrp/partis-zig-core as a zero-dependency drop-in replacement for both the C++ bcrham binary and the C ig-sw binary. Both Zig binaries are selected at runtime via environment variables, so the default C++ path is fully preserved.

Changes

Three hunks in partitiondriver.py (~6 lines total):

• get_hmm_cmd_str: PARTIS_ZIG_CORE_EXE env var overrides the bcrham path when set
• print_partition_dbgfo: guard — Zig binary doesn't emit bcrham timing/debug strings
• check_wait_times: guard — same reason

Two hunks in waterer.py:

• __init__: skip ig-sw binary existence check when PARTIS_ZIG_IGSW_EXE is set
• get_ig_sw_cmd_str: PARTIS_ZIG_IGSW_EXE env var overrides the ig-sw path when set

One new script:

• bin/install-zig-backend.sh: downloads Zig if needed, clones partis-zig-core, builds both static binaries, prints the export lines

Install

bin/install-zig-backend.sh
# copy-paste the two `export PARTIS_ZIG_CORE_EXE=...` and `export PARTIS_ZIG_IGSW_EXE=...` lines printed by the script
partis annotate ...                 # now uses both Zig binaries

No GSL, OpenBLAS, yaml-cpp, or other C++ library dependencies. Works on Linux x86_64 and macOS ARM64 out of the box.

Validation

Correctness: Ran partis cache-parameters on 36 BCR sequences from test/ref-results/test/simu.yaml with both backends:

• Gene calls (V/D/J): identical across all 36 sequences
• Deletion lengths, insertion sequences: identical
• All 26 estimated parameter CSVs: byte-for-bit identical

ig-sw (ksw.zig): Verified on macOS ARM64 — 36/36 sequences pass smith-waterman with the Zig ig-sw, matching the C binary. Note: ksw.zig replaces the SSE2-dependent ksw.c with cross-platform @Vector SIMD, enabling builds on ARM without x86 intrinsics.

Test suite: test/test.py --quick passes with both Zig binaries substituted.

Equivalence: 18 checkpoint-level JSON comparisons between C++ and Zig all pass.

Backward compatibility

Unset both PARTIS_ZIG_CORE_EXE and PARTIS_ZIG_IGSW_EXE → original C++ bcrham and C ig-sw used, no behaviour change.

[matsen]

Verified on Linux x86_64

bin/install-zig-backend.sh
export PARTIS_ZIG_CORE_EXE=.../zig-out/bin/partis-zig-core
export PARTIS_ZIG_IGSW_EXE=.../zig-out/bin/partis-zig-igsw
partis annotate --infname test/ref-results/test/simu.yaml \
  --parameter-dir test/ref-results/test/parameters/simu \
  --outfname /tmp/zig-test-out.yaml

Build succeeded, annotate ran to completion: 36/36 sequences annotated.

[psathyrella]

This looks great! Thanks for working on this.

A few quick thoughts

• I'm not sure I understand why it's worth making the zig binary without debug printing or timing info. The debug printing is really the basis for figuring out what's going on with weird samples and finding bugs in a way that can't be replicated with debuggers. The timing info is ingested by other code so I think is also likely to break things if we change or remove it.
• cache-parameters only runs viterbi, and I didn't run a coverage tool but it seems like since this doesn't run forward or hierarchical agglomeration it probably only uses ~half the ham code. I mentioned elsewhere the test commands we should run, so I'm assuming this is just an initial partial test implementation.
• the packages/ham/ git submodule seems to point to a nonexistent ham commit -- I imagine this is supposed to point to the zig ham repo? Also it looks like ig-sw is still pointing to, well, ig-sw so I think the zig version isn't plugged in yet? In either case I didn't figure out how to have a look through any @copilot actual zig code.
• I assume the environment variable approach to switching is the best for reasons I don't understand, but I would just note that I don't think anything else in partis is configured with environment variables, so it might be good to at least document it carefully, if not switch some other things to use it for more consistency.

[Copilot]

@psathyrella I've opened a new pull request, #340, to work on those changes. Once the pull request is ready, I'll request review from you.

[matsen]

Hey Duncan! Thanks for the comment. I'll work on all these bits.

[matsen]

@psathyrella if it seems like this has legs, I wonder what you would think about bringing the Zig code to be part of the partis repo. It would simplify install and future development.