🤖 perf: Use tsgo for main process compilation (19.6x faster)

[ammar-agent]

Summary

Replaces tsc with tsgo for main process compilation, achieving a 19.6x speedup (3.489s → 0.178s). Completes investigation of tsgo integration opportunities across the entire build toolchain.

Performance Impact

Main process build:

• Before: 3.489s with tsc
• After: 0.178s with tsgo
• Speedup: 19.6x

Type checking (already optimized in PR #260):

• Before: 5.942s with tsc
• After: 0.776s with tsgo
• Speedup: 7.7x

Full build:

• Before: ~26s total
• After: ~22.5s total (~13% faster)

Regression Risk ⚠️

Medium risk - Changing the compiler from tsc to tsgo for production builds:

1. TypeScript 7 preview: @typescript/native-preview is pre-release software (v7.0.0-dev)
2. Different code generator: While type checking is identical, the emitted JavaScript may have subtle differences
3. Platform-specific binaries: Relies on Go-based native binaries for darwin/linux/win32 across x64/arm64
4. Path alias resolution: Still requires tsc-alias post-processing (tsgo doesn't resolve @/ imports)

Mitigation:

• CI tests run on both macOS and Linux
• Integration tests verify IPC communication paths
• E2E tests verify full application behavior
• Build is deterministic (no conditional branches)

Rollback plan: If issues arise, revert to tsc by changing one line in Makefile (line 77).

Changes

• Updated Makefile build-main target: Uses tsgo directly, no fallback branches
• Updated Makefile typecheck target: Removed tsc fallback (already had branch from PR 🤖 feat: Replace tsc with tsgo for 7.7x faster type checking #260)
• Added build reproducibility guidelines: Documents why branches are avoided in build targets
• Requires tsgo: Builds fail fast if tsgo missing (in devDependencies, always available)

Investigation Findings

Analyzed every component of the build toolchain to maximize tsgo usage:

Component	Current Tool	Can Use tsgo?	Status
Main process build	tsc	✅ Yes	✅ Implemented
Renderer build	Vite+esbuild	❌ No	N/A (Vite architecture)
Preload build	bun	❌ No	N/A (already fast)
Type checking	tsgo	✅ Yes	✅ Done (PR #260)
Dev watch (main)	tsgo -w	✅ Yes	✅ Done (PR #260)

Key findings:

1. Path alias resolution: tsgo doesn't resolve @/ imports, so tsc-alias still required (~0.7s overhead, acceptable)
2. Vite can't use tsgo: Different compilation model (esbuild + HMR + bundling)
3. Main build time dominated by Vite: ~21s for renderer vs ~0.2s for main process

Conclusion: We've maximized tsgo usage. The remaining build time (~21s Vite renderer compilation) cannot be optimized with tsgo due to architectural constraints.

Build Reproducibility

Removed all conditional branches (if/else fallbacks) from build targets. Builds now fail fast with clear errors if dependencies are missing, rather than silently using different compilers.

Added to Makefile header:

AVOID CONDITIONAL BRANCHES (if/else) IN BUILD TARGETS AT ALL COSTS.
Branches reduce reproducibility - builds should fail fast with clear errors if dependencies are missing, not silently fall back to different behavior.

Testing

• ✅ Full build succeeds: make build
• ✅ Type checking works: make typecheck
• ✅ No unresolved @/ imports in compiled output
• ✅ Dev watch works: make dev
• ⏳ CI checks (static, integration, e2e) pending

Generated with cmux

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.