Cache formula lookups during dependency expansion

[p-linnane]

During a single dependency expansion pass, Dependency.expand resolved every dependency edge with a fresh dep.to_formula, and FormulaInstaller#expand_dependencies_for_formula re-ran dep.satisfied? for equal edges. Homebrew's factory cache is disabled on the install/upgrade path, so each to_formula reloads the formula, and a shared low-level dependency reached through many edges is loaded once per edge. That is the repeated Formulary::FromAPILoader: loading work reported in #22896, worst on Linux when the implicit bubblewrap dependency is unsatisfied and pulled into many subtrees.

This adds two caches scoped to one expand_dependencies_for_formula call and discarded afterwards: a formula cache so equal edges reuse one resolved Formula, and a satisfied cache so equal edges check dep.satisfied? once. Both key on the full dependency identity via hash/eql? (including UsesFromMacOSDependency bounds), so subclass state is never collapsed. Behaviour is unchanged: the caches only remove repeated work, and formula_cache raises unless a cache_key is also given, so it is only usable where expansion results are already memoized.

---

• Have you followed our Contributing guidelines?
• Have you checked for other open Pull Requests for the same change?
• Have you explained what your changes do? Performance claims (e.g. "this is faster") must include Hyperfine benchmarks.
• Have you explained why you'd like these changes included, not just what they do?
• For bug fixes, have you given step-by-step brew commands to reproduce the bug?
• Have you written new tests (excluding integration tests)? Here's an example.
• Have you successfully run brew lgtm (style, typechecking and tests) locally?

---

• AI was used to generate or assist with generating this PR.

Claude Code (Opus 4.8) implemented the per-expansion caching and its tests and reviewed the change for correctness. I reviewed the diff, confirmed the new specs fail without the caches and pass with them, and ran brew lgtm plus targeted brew tests --only=dependency_expansion / --only=formula_installer.

---

[Copilot]

Pull request overview

This PR improves dependency expansion performance by avoiding repeated dep.to_formula loads and repeated dep.satisfied? checks during a single expansion pass, addressing the Linux slowdown described in #22896 where formula factory caching is disabled on install/upgrade paths.

Changes:

• Add an optional per-expansion formula_cache to Dependency.expand and use it for recursive expansion to reuse resolved Formula objects for equal dependency edges.
• Add per-expansion caching in FormulaInstaller#expand_dependencies_for_formula so equal dependency edges only run dep.satisfied? once per expansion call.
• Add unit tests covering both caching behavior and ensuring UsesFromMacOSDependency bounds remain part of dependency identity (no incorrect cache collapsing).

Reviewed changes

Copilot reviewed 4 out of 4 changed files in this pull request and generated no comments.

File	Description
Library/Homebrew/dependency.rb	Adds optional formula_cache: support to reuse dep.to_formula results during expansion (guarded by cache_key).
Library/Homebrew/formula_installer.rb	Introduces per-expansion formula_cache and satisfied_cache to cut redundant work in installer dependency expansion.
Library/Homebrew/test/dependency_expansion_spec.rb	Adds tests proving formula caching works for equal dependencies and does not collapse differing uses_from_macos bounds.
Library/Homebrew/test/formula_installer_spec.rb	Adds tests ensuring dep.satisfied? is evaluated once per equal edge and keeps uses_from_macos bounds distinct.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[MikeMcQuaid]

Thanks again @p-linnane!