backend/golang: simplify runtime minimization design and track follow-up

[emil14]

Context

We evaluated a Datadog-inspired optimization for Go binary size: include only used runtime funcs in generated executables instead of embedding the full internal/runtime/funcs set.

What was prototyped

• Collect used runtime refs from IR (FuncCall.Ref).
• Parse runtime/funcs/registry.go and runtime/funcs/*.go.
• Build mapping ref -> creator -> source file.
• Build transitive file dependency closure via Go AST symbol usage.
• Copy only selected runtime func files and generate minimal runtime/funcs/registry.go.

Result

• The prototype can significantly reduce small-program binary size/deps (example from PR prototype: ~6.56MB -> ~2.01MB for a minimal case).
• However, implementation complexity became too high for baseline compiler/backend flow and too hard to review/maintain in its current form.

Why we should defer this from baseline PRs

This optimization currently mixes abstraction layers too tightly:

• Neva-level runtime refs from IR,
• Go runtime registry internals,
• Go AST file-level dependency analysis.

This cross-layer coupling increases cognitive load and maintenance risk.

Follow-up design directions (to simplify)

1. Move runtime funcs toward clearer package boundaries (currently mostly naming-based grouping).
2. Adopt a stronger convention: one runtime func creator per file (or per very small cohesive file set).
3. Consider making runtime func refs structured (entity reference), not free-form string only.
4. Use package/file-level selection driven by structured refs first; keep AST dependency analysis as optional/secondary.
5. Keep build-path improvements (e.g. -trimpath, -buildvcs=false, shared build args) separate from deep minimization logic.

Measurement and observability

Follow-up work should include repeatable measurements and regression tracking; related meta issue: #1015.

Status

• Keep current PR scope minimal (build flags + small safe refactors).
• Continue runtime minimization as a dedicated design/implementation track.

[emil14]

AgentOps check after webhook sync: please provide an implementation plan and acceptance criteria.

[emil14]

Retry after timeout tuning: please post concise plan + acceptance criteria only (no deep repo scan).

[emil14]

Second retry after strict planning guard: please answer with concise plan + acceptance criteria only.

[emil14]

---

Next Step: This plan outlines a phased approach to simplify runtime minimization while preserving the core benefit of binary size reduction. The focus is on establishing clear boundaries first, then implementing simpler selection logic, and finally adding measurement and integration.

The plan respects the concern about cross-layer coupling and maintains separation between compiler layers. Once approved, we can proceed to implementation.

opencode session  |  github run