fix(cargo-gamma): support binary-only package selections

[chinayangxiaowei]

Problem

cargo gamma run cannot complete when every selected package has only binary targets. Schema convergence always runs cargo check --lib --bins --tests; Cargo rejects --lib with no library targets found in package before any mutant can be measured. A mixed selection containing at least one library package already works with those selectors.

Change

• Omit --lib when every selected package is binary-only.
• Retain the existing --lib --bins --tests command for mixed selections. No manifest or source rewrite is needed.
• Cover binary-only and mixed selections, including a library without a test harness. Update the design and CLI documentation.

Validation

• cargo fmt --all -- --check passed.
• A direct Cargo 1.97.1 check of a two-member fixture confirmed that --lib --bins --tests --workspace succeeds, while selecting only its binary package fails with no library targets found.
• On macOS 15.7.2, the new convergence and end-to-end session tests passed with the macOS process cleanup changes from fix(cargo-gamma): handle macOS process group cleanup and CLI builds #240 applied. The convergence test fails on the unmodified upstream base at the existing macOS process cleanup assertion addressed by fix(cargo-gamma): handle macOS process group cleanup and CLI builds #240.
• End-to-end macOS tests used TMPDIR=/private/tmp to avoid the host /var to /private/var path alias in the existing scratch-directory check.
• crates/cargo-gamma-lib/docs/design/system-tests.md designates the existing tests/session.rs cases as real-toolchain acceptance tests. The two new session cases intentionally exercise Cargo's target-selector acceptance and feature unification; scripted command responses cannot establish those properties.

This PR contains only the binary-only selection fix and a mixed-workspace regression test. It does not include #240.

[Copilot]

🟡 Changes recommended

The rollback regression test does not verify that installation reaches the intended second-package failure.

1 open finding

• Test does not verify rollback after second-package failure · New

What changed in this PR

Fixes schema convergence for binary-only and mixed Cargo workspaces.

Changes:

• Selects convergence targets based on library availability.
• Adds and restores temporary library targets for mixed workspaces.
• Adds regression tests and updates design documentation.

File	Description
crates/​cargo-gamma/​README.md	Documents binary-only workspace support.
crates/​cargo-gamma/​docs/​DESIGN.md	Describes mixed-workspace convergence.
crates/​cargo-gamma-lib/​tests/​session.rs	Adds end-to-end workspace tests.
crates/​cargo-gamma-lib/​src/​exec/​measure.rs	Passes library metadata to convergence.
crates/​cargo-gamma-lib/​src/​exec/​build/​tests.rs	Tests temporary-library behavior.
crates/​cargo-gamma-lib/​src/​exec/​build/​temporary_libraries.rs	Implements temporary targets and restoration.
crates/​cargo-gamma-lib/​src/​exec/​build.rs	Selects commands and temporary targets.
crates/​cargo-gamma-lib/​src/​discover/​survey.rs	Records packages with libraries.
crates/​cargo-gamma-lib/​docs/​IMPLEMENTATION.md	Documents implementation details.
crates/​cargo-gamma-lib/​docs/​DESIGN.md	Updates the convergence contract.

🧠 Review effort: Balanced

---

💡 Configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

[Copilot]

🟡 Changes recommended

Synthetic libraries can fail valid workspaces that deny unused dependency warnings.

2 open findings

• Exempt the synthetic target from unused dependency lint · New
• Inject a process runner instead of invoking real subprocesses · New

1 resolved since last review

• Test does not verify rollback after second-package failure

🧠 Review effort: Balanced

[Copilot]

🟢 Approval recommended

The selector logic, failure-safe restoration, metadata plumbing, documentation, and regression coverage are consistent and complete.

2 open findings

• Exempt the synthetic target from unused dependency lint
• Inject a process runner instead of invoking real subprocesses

🧠 Review effort: Balanced

[Copilot]

🟢 Approval recommended

The selector logic matches the documented contract and is covered by focused unit and real-toolchain acceptance tests.

2 open findings

• Exempt the synthetic target from unused dependency lint
• Inject a process runner instead of invoking real subprocesses

🧠 Review effort: Balanced

[Copilot]

🟢 Approval recommended

The selector logic matches the documented contract and is covered across binary-only and mixed-package scenarios.

0 open findings

2 resolved since last review

• Exempt the synthetic target from unused dependency lint
• Inject a process runner instead of invoking real subprocesses

🧠 Review effort: Balanced

[Copilot]

🟢 Approval recommended

The selector logic, metadata propagation, documentation, and regression coverage consistently address binary-only and mixed selections.

0 open findings

🧠 Review effort: Balanced

[Martin Taillefer (geeknoid)]

No additional findings in the current selector-only fix. Library presence comes from Cargo target metadata rather than test-harness availability, and the check examines the selected package set. All-binary selections omit --lib; mixed selections keep the existing combined selectors. The synthetic-library/manifests mechanism discussed earlier is gone.

The fixtures cover a binary-only crate, a mixed workspace with a test-enabled dependency feature, a library without a harness, and selecting only the binary member. These are appropriately documented as real-toolchain system acceptance tests rather than assumed unit tests.

All five passes covered metadata collection, preflight-to-converger wiring, target selection, tests, and contracts. Generated README output was excluded as an independent review surface. No builds, tests, or mutation runs were executed.

[Codecov Comments Bot (codecov-commenter)]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 97.8%. Comparing base (0e9a995) to head (5758b27).

Additional details and impacted files

@@           Coverage Diff            @@
##            main    #241      +/-   ##
========================================
- Coverage   98.9%   97.8%    -1.1%     
========================================
  Files        304     118     -186     
  Lines      44469   18965   -25504     
========================================
- Hits       44008   18561   -25447     
+ Misses       461     404      -57     

Flag	Coverage Δ
linux	97.8% <100.0%> (-1.1%)	⬇️
linux-arm	97.8% <100.0%> (-1.1%)	⬇️
scheduled	?
windows	97.6% <100.0%> (-1.3%)	⬇️

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ 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.

[Copilot]

🟢 Approval recommended

The selector logic matches the documented contract and is covered across binary-only, mixed, and library-without-harness cases.

0 open findings

🧠 Review effort: Balanced

[Copilot]

🟢 Approval recommended

The selector logic matches Cargo package scope and is covered for binary-only, mixed, and no-harness library cases.

0 open findings

🧠 Review effort: Balanced