test(globals): cover parse_args, the thread/RAM scheduler, and the print_* formatters (#182)

[TimD1-bot]

Note

Authorship: the content below was drafted by Claude Opus 5 (an AI coding agent) and
filed via gh under @TimD1-bot, a bot account operated by @TimD1. It reflects the
agent's analysis, not a statement authored by @TimD1.

Adds 96 cases to tests/unit/src/test_globals.cpp, taking Globals::parse_args from zero unit
coverage to the whole CLI contract. No source file is modified. The suite goes from 561 to
657 passing cases.

What's covered

parse_args — 83 cases

Group	Cases
argc < 4 short-circuit	no arguments, -h, --help, -v, --version, -ci, and an unrecognized argument — all exit(0)
Mandatory positionals	all three accepted; each of the three open failures; optional-before-mandatory
Verbosity pre-pass	each valid level, non-numeric, and both out-of-range ends
Per flag	-b, -p, -f, -l, -sv, -q, -mq, -n, -x, -o, -e, -i, -s, -t, -ct, -r — a happy case each, plus missing-value, bad-numeric, and bound-violation wherever the parser enforces one
Main loop, non-exiting	-h, --version, -ci — none of these exit, unlike their argc < 4 counterparts
Unrecognized / consumed	--bogus errors; -v <n> is consumed without effect and the flag after it still parses
Cross-field, post-loop	--max-qual below --min-qual; --largest-variant below --sv-threshold; --max-supercluster-size below --largest-variant + 2; --largest-variant 1

set_thread_ram_steps — 7 cases, driven directly rather than through an argv array, since
both the constructor and the end of parse_args call it: the 64-thread ladder, a single thread,
integer halving on a non-power-of-two (10 → {10, 5, 2, 1}), the RAM-per-thread division, that
repeated calls replace rather than append, thread_nsteps agreeing with both vector sizes, and
that an out-of-range thread count yields no steps rather than dividing by zero.

print_version, print_usage, print_citation — 6 cases on captured stdout, including
that the usage text interpolates the live settings rather than compiled-in literals, and that it
omits the four flags whose usage lines are commented out at src/globals.cpp:492-499.

Harness notes

Three pieces, all local to the test file:

• ArgsFixture writes an openable query and truth VCF into a TempDir and points the
  reference at the committed tiny.fasta, so each case states only the flags it varies.
  parse_args opens argv[1] and argv[2] with bcf_open and argv[3] with fopen, so every case
  reaching the optional-argument loop needs three files that exist.
• parse() copies each argument into its own mutable buffer, since parse_args takes
  char ** rather than const char **. It also closes the reference FASTA, which parse_args
  opens and never closes — 96 cases would otherwise leak one descriptor each.
• parse_showing_stdout() dup2s stdout onto stderr inside the death-test child. A
  death-test matcher only sees the child's stderr, but print_usage, print_version and
  print_citation write to stdout; without this the argc < 4 group could assert only an exit
  code, which makes -v, -h and -ci indistinguishable. I confirmed the matchers actually
  bite by temporarily patching one to a sentinel string and watching the case fail.

Defects found — pinned, not fixed

Three cases document current behavior rather than enforcing it. Each is marked as such in-file
and names the issue that will change it, so the amendment is a grep away:

• parse_args: a trailing comma in --filter adds an empty filter name #199 — a trailing comma in --filter appends an empty filter name, which then warns once
  per callset from src/variant.cpp:790-793. Pinned by ParseArgs.FilterTrailingComma.
• parse_args: -b's catch block is unreachable for an unopenable BED and mislabels a malformed one #200 — -b's catch block is unreachable for an unopenable BED (the bedData
  constructor exits first), and for its one reachable trigger it reports an invalid filename
  when the contents were invalid. Both routes are pinned, by ParseArgs.BedBadFileErrors and
  ParseArgs.BedMalformedLineErrors.
• Unfiled — the missing-value error for --verbosity at src/globals.cpp:92 is
  unreachable: the pre-pass loop is for (int i = 0; i+1 < argc; i++), so i never exceeds
  argc-2 and i == argc cannot hold. A trailing -v is never examined, the main loop then
  advances past the end, and the parse succeeds with verbosity unchanged instead of reporting the
  missing value. Pinned by ParseArgs.VerbosityMissingNotReached.

-i/--max-iterations and -s/--max-supercluster-size carry in-file comments naming #46 and
#47, since those issues remove and rename them respectively.

Verification

• cd tests/unit/build && make — clean; the only warning is the pre-existing unused-parameter
  one at src/dist.cpp:1063.
• ./test_vcfdist — 657 passed, 74 suites.
• cd tests && pytest — 65 passed.

Closes #182