test: DRY up the suite with shared builders + parametrized tables

[FBumann]

What

Two passes over the test suite to make it DRYer and more scannable.

Pass 1 — shared fixture builders

Every module reimplemented the same synthetic pytest-benchmark shapes under different names. tests/_builders.py now owns three primitives:

• write_run(path, benchmarks) — write a {"benchmarks": [...]} run file.
• bm(name, *, t, peak, allocations, total_bytes, rss, params) — one benchmark entry.
• blob(peak, *, allocations, total_bytes, rss) — a benchmem per-repeat series.

Every module migrated to them (one commit each). Scenario-specific builders stay local.

Pass 2 — parametrized tables + assertion helpers

Collapsed the clone-families (each was N near-identical functions) into parametrized tables, and added output helpers:

• test_plotting.py: axis-scale (6→1), band-whisker (4→1), where-filter (2→1) tables.
• test_cli.py: --fail-on gate (4→1), plot bad-flag (5→1), flamegraph usage (4→1) tables; plus capture_view/view_calls fixtures replacing ~9 hand-rolled monkeypatch stubs.
• test_compare.py: render(runs, **kwargs) -> str helper replacing ~20 StringIO()/compare_runs(out=)/getvalue() triads; unknown-option rejection table (4→1).

Result

• -183 lines across the test files (462 insertions, 645 deletions) plus the ~70-line builders module.
• 284 → 260 test functions, but the collapsed ones became parametrized cases, so coverage is unchanged: 309 passed, identical to main.
• Each commit is scoped to one file / one idea, and pre-commit (ruff + mypy) is green on every commit.

Not included

Grouping test_cli.py into classes — the # --- section --- dividers already give grouping and classes would add churn without functional gain. Easy follow-up if wanted.

🤖 Generated with Claude Code

[read-the-docs-community]

Documentation build overview

📚 pytest-benchmem | 🛠️ Build #33391293 | 📁 Comparing c918e5b against latest (3da9bb1)

  🔍 Preview build  

2 files changed

± changelog/index.html
± compare-plot/index.html