test/coverage: authoritative full-suite run 3 — 68.0% line / 50.2% branch

[gburd]

Authoritative full-suite coverage run 3

Fresh, authoritative code-coverage measurement on a dedicated AWS EC2
c7i.24xlarge (96 vCPU, AL2023), master @ v5.3.32 (27f8e7b6c), then the
instance was terminated. Measurement/report only — no engine code changed.

ONE instrumented build (gcc 11.5, -O0 -g --coverage), .gcda accumulated
across the full parallel Tcl suite AND every COV_* block, merged in a
single lcov --capture --directory .libs.

Headline (src/-only, dbinc_auto excluded, branch coverage on)

metric	run 3	interim (run 2)	authoritative (v5.3.30-era)
line	68.0% (49192/72377)	62.3%	48.0%
branch	50.2% (39238/78236)	46.2%	36.1%
function	80.5% (2129/2644)	72.3%	—

+20.0 pts line / +14.1 pts branch vs the authoritative baseline;
+5.7 / +4.0 vs the interim. Wall time ~89 min.

What ran

• Full access-method matrix (btree/hash/recno/queue/heap, ~135 tests each, +verify/salvage)
• All subsystem groups (env/lock/log/memp/mutex/txn/sdb/sec/compact/partition/...)
• Replication: 22 rep + 15 repmgr = 37 PASS (hangers rep016/034, repmgr024/026 skipped)
• recd group + 26 curated recd recovery-handler tests
• COV_* blocks: curated subset (crypto sec001/002, statprint, mvcc, lock007, test143),
  deadlock+DB_REGISTER drivers, and the C drivers (backup, os_aio, recd_compact PASS)

Honest caveats (see FULL-COVERAGE-REPORT-3.md)

1. xa=0% in this capture. liburing-devel was missing at build time, so
   xa_direct failed to link (-luring). After installing it the XA driver
   re-ran and PASSED, but the EC2 instance was reclaimed (went to
   shutting-down on its own) before the re-capture could be pulled back — so
   the verifiable headline is 68.0% with xa=0%; with XA counted it's ~68.3%.
   Next round: dnf install liburing-devel up front.
2. db_upgrade PARTIAL — real paths ran; the driver's synthetic hash-fixture
   aborts on a spares-array bug (run_upgrade.sh, not the engine).
3. os_aio io_uring backend cold — io_uring_queue_init returned ENOSYS on
   the host (kernel io_uring disabled). Other AIO backends covered.

Contents

• test/coverage/full_run3_combined.sh — the combined driver
• test/coverage/FULL-COVERAGE-REPORT-3.md — report + per-subsystem + top cold files + path-to-80%
• test/coverage/full-run-3/ — cov-src.info, ranking, subsystems, summary
• test/coverage/baseline.txt — ratcheted 62.3/46.2 → 68.0/50.2

[gburd]

Superseded: the coverage report + baseline ratchet (68.0/50.2) were merged via #94, and the uniquely-valuable full_run3_combined.sh + cov-subsystems.txt are merged via #96. The remaining diff in this branch is stale duplicate content (db_join/db_method from #92, recd_handlers from #93, fi_sweep from #90 — all already on master). Thanks — the run itself produced the authoritative number.