Tracking: open PRs that conflict with main

This is a maintainer tracking list, not a task for a newcomer. The open items are the unchecked PRs (#282, #278, #237, #213), which the status notes say are left to the maintainer, and #146 versus #295, which needs a decision on which supersedes the other before rebasing. Reproduce a conflict locally with git merge-tree --write-tree origin/main <pr-head>. Done means the PR merges cleanly, and fmt, clippy -D warnings and the test suite pass.

Snapshot of open pull requests that no longer merge cleanly into main (abf2ab7), checked locally with git merge-tree --write-tree origin/main <pr-head> for each of the 45 open non-dependabot PRs. Review order and stacking are discussed in #195; this issue only tracks the merge conflicts.

22 of them conflict. Most of the churn comes from three merges: the bounded-memory BAM sort (#214, in #290), the batch-1 fixes (#289), and the retirement of CLAUDE.md (#251).

Conflicts in source code (needs a real rebase)

• #282 fix(chimeric): well-formed single-end WithinBAM records with STAR's hard-clip rules: src/chimeric/detect.rs, src/chimeric/output.rs, tests/alignment_features.rs
• #278 perf(io): encode SAM/BAM records on the align workers (stacked on #276): src/io/bam.rs, src/io/sam.rs, src/lib.rs
• #276 perf(io): parallel order-preserving BGZF compression for BAM output: src/io/bam.rs
• #272 feat(io): detect input compression by magic bytes, optional bz2/zstd/xz: Cargo.toml, DIVERGENCE.md, src/io/fastq.rs
• #271 fix(align): pair-subset dedup requires same chromosome and strand (stacked on #254): CHANGELOG.md, src/align/read_align.rs
• #254 fix(align): a soft-clipped variant of a pair is not a second locus: CHANGELOG.md, src/align/read_align.rs
• #237 feat: output anndata: Cargo.lock, src/junction/sj_output.rs, src/solo/count.rs, src/solo/mod.rs
• #225 feat(io): optional pure-Rust parallel gzip input decoding (rapidgzip-core): CHANGELOG.md, Cargo.toml, src/io/fastq.rs
• #213 feat(io): native CBQ input support: Cargo.toml, src/io/fastq.rs, src/lib.rs, src/params/mod.rs, src/solo/whitelist.rs
• #160 genome: --genomeTransformOutput SAM reports original coordinates: CHANGELOG.md, src/chimeric/detect.rs, src/params/mod.rs, tests/alignment_features.rs
• #146 aligner core: annotated-junction stitching, alignEndsType, in-recursion length penalty: src/align/score.rs, src/align/seed.rs, src/align/stitch.rs, src/chimeric/detect.rs, src/junction/gtf.rs, tests/parameter_surface.rs
• #109 genomeGenerate: libsais suffix-array builder, selected by --limitGenomeGenerateRAM: Cargo.toml, src/lib.rs

Conflicts only in CHANGELOG, Cargo manifests, docs or tests

These are usually mechanical: keep both sides.

• #275 feat(solo): optional CellRanger v3 .h5 count matrices via hdf5-pure: Cargo.toml, tests/alignment_features.rs
• #273 feat(io): optional paraseq FASTQ backend: Cargo.toml
• #246 docs: record dependency decisions in DEPENDENCIES.md: Cargo.toml
• #241 fix(solo): port STAR's actual CellRanger4 clip rules: CHANGELOG.md
• #239 feat(cli): parameter files, genome version check, and a shell-run readFilesCommand: tests/parameter_surface.rs
• #232 feat(degnorm): DegNorm-style transcript degradation normalization: CLAUDE.md, DIVERGENCE.md
• #198 integrate hyalite v0.2.0 for CellRanger4: Cargo.toml, tests/alignment_features.rs
• #161 genome: --genomeType SuperTranscriptome: CHANGELOG.md, tests/alignment_features.rs
• #148 clip: CellRanger4 3' poly-A trim: tests/alignment_features.rs, tests/parameter_surface.rs
• #116 feat: STARlong binary + long-read window-coverage filter: tests/parameter_surface.rs

Already rebased

• #226 (Phase 14.7): merged main. Because #214 replaced the in-memory sort, CB/UB are now filled as records leave the sort.

Merge clean today: #297, #296, #295, #294, #293, #283, #274, #255, #253, #248, #243, #226, #184, #182, #179, #178, #176, #174, #165. Some of them are stacked and only clean against their own base.

Status (2026-10-09)

main was merged into 11 of these branches (checked boxes above), each with fmt, clippy -D warnings and the full test suite passing. A few needed more than conflict markers:

• #232: --quantMode GeneCoverage added to the --quantMode validation main gained in #283; the DegNorm section of DIVERGENCE.md renumbered to §4.4 (main's §4.3 is the sort budget, referenced from code).
• #198: an earlier merge on that branch had dropped main's test_run_mode_solo_cell_filtering_calls_cells_from_a_raw_matrix; restored.
• #116: main added a junction-scan cache argument to stitch_align_to_transcript; the long-read stitcher now passes one.
• #254: with main merged, 58 mates fixed and 0 newly differing on 200k yeast pairs.

The remaining 11 (#282, #278, #276, #272, #271, #237, #225, #213, #160, #146, #109) are left to the maintainer. Note on #146: it overlaps heavily with #295 and the parity stack (#301, #302, #304) in stitch.rs/score.rs/seed.rs, so it is worth deciding which approach supersedes the other before rebasing it.

Status (2026-10-09, after main moved to b5a70d0, #307)

Re-checked every open PR against the new main and merged it into all that conflicted, except the ones below. Each merge passes fmt, clippy -D warnings and the full test suite (with the PR's cargo feature when it has one).

• #295, #301, #302, #304 (parity stack): main merged into #295 and propagated; the stack still fixes 350 mates and breaks none against the new main.
• #293, #254, #241, #271: CHANGELOG only.
• #272: its magic-byte detector already used MultiGzDecoder (main's #217 fix); main's multi-member test kept. DIVERGENCE.md section renumbered §4.4 (note: #232's DegNorm section is also §4.4; whichever lands second needs §4.5).
• #225: rapidgzip path kept with main's MultiGzDecoder fallback.
• #109: main's "limitGenomeGenerateRAM accepted but not enforced" warning dropped since the branch uses it; the yeast index it builds is byte-identical to STAR's.
• #160: --genomeTransformOutput SAM now also refuses --chimSegmentMin > 0 (the PR never mapped chimeric output back; main's #307 WithinBAM would have mixed coordinate spaces).
• #276: main's external sort with #276's parallel BGZF (spill runs too); records md5-identical to main, including a 331-run spill.
• #146: 239 mates fixed, 0 newly differing on 200k yeast pairs with main merged. Overlaps #295 (both add sj_a to window/exon blocks and stitch annotated junction halves directly) and contradicts it on three points: the sj_a type, whether the short-exon early rejection is skipped for junction halves, and how annotated non-canonical junctions are shifted back. Only #295 drops covered PE sub-alignments; only #146 has alignEndsType, the in-recursion length penalty, the mismatch-over-read cap, alignInsertionFlush and flagDirMap. One of the two should be rebased on the other.

Still open: #278 (being merged onto the updated #276), #237, #213. #294 is superseded: main's #307 already makes paired-end nM the pair total, which is all #294 does.