fix(sort): scale the per-thread memory budget by the sort phase

[tfenne]

Stacked on #691 — please review/merge that one first; this PR targets its branch, so the diff here is just the memory change.

--memory-per-thread multiplied --max-memory by --threads alone. Since --sort-threads defaults to --threads and not the reverse, a run that set only the per-phase override got the one-thread budget while sorting with eight workers:

$ fgumi sort -i in.bam -o out.bam --max-memory 100M --sort-threads 8
INFO  Max memory: 95.4 MiB (95.4 MiB/thread x 1 threads)
INFO  Threads: sort 8, merge 1

The budget sizes the in-memory accumulation buffer, which the sort phase fills, so it now scales by that phase:

INFO  Max memory: 762.9 MiB (95.4 MiB/thread x 8 threads)

Why max(--threads, --sort-threads) and not just --sort-threads

Scaling strictly by the sort phase would be a silent regression for the pattern --sort-threads exists to serve. -@ 32 --sort-threads 4 is the documented way to cede cores to an upstream producer while keeping the merge wide; under sort-phase-only scaling that run's buffer drops from 32× to 4× — an eightfold cut, trading a scheduling hint for a throughput cliff. Taking the larger of the two raises the budget exactly where --threads under-counts and never lowers it, so no existing invocation resolves less memory than it does today.

Scope

• --threads 0 still reaches resolve_memory_budget's rejection rather than being clamped up by the new helper.
• --memory-per-thread false is unaffected (still a fixed total).
• The Max memory log line and the --max-memory auto initial-capacity cap both follow the same count, so neither can disagree with the resolved budget.
• Only fgumi sort is touched. QueueMemoryOptions (filter/clip/correct) has a single thread count with no sort/merge phases, so it needs no equivalent — and the README/performance-tuning memory notes describe that pipeline-queue path, not this one.

The helper re-derives the sort-phase fallback because the budget is needed before the sorter exists (so it can't read phase1_threads off it, as the Threads: line in #691 does). A test asserts the two definitions agree, so they can't drift.

Tests: 7 cases over the multiplier (including both zero cases), 4 drift-guard cases, and two end-to-end tests asserting the logged budget — one for the bug, one pinning the -@ 32 --sort-threads 4 shape against the regression above. I confirmed the end-to-end tests fail before the fix, at 95.4 MiB (... x 1 threads).

ci-fmt/ci-lint clean. ci-test is 6905 passed / 3 failed — the same test_input_source_matrix::declared_{sam,stdin}_support_* trio that fails on a pristine main (they diff R-generated simplex_qc.pdf byte-for-byte and R stamps /CreationDate into it), as noted in #690 and #691.

[coderabbitai]

Note

Reviews paused

Use the following commands to manage reviews:

• @coderabbitai resume to resume automatic reviews.
• @coderabbitai review to trigger a single review.

Use the checkboxes below for quick actions:

• ▶️ Resume reviews
• 🔍 Trigger review

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}

[nh13]

@coderabbitai pause

[coderabbitai]

✅ Action performed

Reviews paused.

[codecov]

Codecov Report

❌ Patch coverage is 96.55172% with 1 line in your changes missing coverage. Please review.
✅ Project coverage is 93.93%. Comparing base (521e88d) to head (a628610).
⚠️ Report is 1 commits behind head on main.

Files with missing lines	Patch %	Lines
src/lib/commands/sort.rs	96.55%	1 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #692      +/-   ##
==========================================
- Coverage   93.95%   93.93%   -0.02%     
==========================================
  Files         178      178              
  Lines      108086   108109      +23     
==========================================
+ Hits       101552   101553       +1     
- Misses       6534     6556      +22     

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.