perf(go): cache lowering metadata

[akonwi]

Summary

• precompute and cache generated Go declaration names and top-level collision sets during lowering
• replace repeated import-alias collision scans with constant-time lookups
• carry foreign Go result ABI shape from checker through AIR, avoiding duplicate go/packages loading in the backend
• reject missing foreign result-shape metadata instead of silently treating it as a direct result

Performance

Warm build measurements:

Project	Before	After
migrate	~13.9s	~1.6s
maestro/server	~21.6s	~2.7s
tinear	~19.0s	~2.3s

Profiling showed repeated generated-name/import-collision scans dominating Go AST lowering. Rendering, workspace I/O, and final warm Go compilation were comparatively small.

Correctness

• preserves existing type, function, global, and enum-variant naming behavior
• updates collision planning as synthetic Maybe types are introduced
• propagates direct, (T, error), error, and (T, bool) foreign result shapes through generic functions, methods, method values, and function values
• keeps unknown result metadata explicit through checker and AIR so Go lowering fails clearly

Validation

• cd compiler && go test ./... -count=1
• cd compiler && go vet ./...
• git diff --check
• successful builds of Migrate, Maestro, and Tinear
• Migrate runtime smoke test
• randomized naming parity and generic FFI differential checks during development
• final code review completed with the missing-result-shape finding addressed and re-reviewed

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
ard-lang-dev	Ready	Preview, Comment	Jul 21, 2026 3:25pm