Investigate migrating formatter from Prettier to oxfmt (oxc formatter)

[timotheeguerin]

Summary

Proof-of-concept investigation into replacing Prettier with oxfmt (the oxc formatter) for the monorepo. Result: large speed win with low churn for the JS/TS/JSON bulk. oxfmt can't format a few file types, so the proposal is a hybrid where oxfmt owns the bulk, tsp format owns .tsp, and Prettier is kept only for .astro + shell.

Tested with oxfmt 0.59.0 (npm hybrid build).

Key results

• ~30× faster on the JS/TS/JSON bulk: oxfmt --check = ~2.7s on 2947 files vs Prettier format:check = ~80–100s.
• Low churn: only 55 / 2947 files (1.9%) differ — genuine formatting deltas (union-type line-breaking, arrow-arg wrapping), not import reordering.
• No JSON / YAML / CSS churn — byte-identical to Prettier.
• .prettierrc migrated cleanly via oxfmt --migrate=prettier (all core options + every .prettierignore pattern). Only the 4 plugins are dropped (expected).

Coverage

oxfmt natively handles: JS/TS/TSX, JSON, YAML, Markdown, CSS, GraphQL, TOML.

Gaps (must live elsewhere):

File type	Count	Plan
.tsp	356	tsp format (compiler's own formatter; the prettier plugin just wraps it)
.astro	47	Unsupported in 0.59 → residual Prettier
.sh	1 (packages/standalone/install.sh)	No shell support → residual Prettier

Benchmarks (M-series mac, 10 threads)

Step	Files	Wall time
Prettier format:check (current)	~all	~80–100s
oxfmt --check .	2947	~2.7s
tsp format --check	324	~1.8s
Prettier residual (*.{astro,sh})	48	~4.1s
Proposed combined pipeline	—	~9s

Churn detail (the 55 files)

26 .ts, 25 .tsx, 4 .md. Representative diffs:

• Union types that fit within printWidth are broken to one member per line with a leading | by oxfmt, where Prettier 3.9.5 keeps them on one wrapped line.
• arr.map((x): T => ({...})) — oxfmt wraps the callback argument differently.

One-time reformats; the resulting style is stable and self-consistent.

⚠️ Do NOT enable sortImports: it re-sorts with the perfectionist algorithm (≠ prettier-plugin-organize-imports order) and churn jumps to 1337 files (45%).

Import organizing — behavioral gap

• Current prettier-plugin-organize-imports sorts + removes unused imports.
• oxfmt sortImports is off by default; when on it only sorts (different order), never removes unused.
• oxlint already has no-unused-vars: error, which flags unused imports (blocks CI) but does not auto-remove them on --fix.
• Recommendation: keep sortImports off. Either (a) drop import-organizing and rely on oxlint, or (b) keep prettier-plugin-organize-imports as a residual TS/JS step. (a) is cleaner; (b) preserves exact current behavior.

Proposed end-state

• oxfmt owns: JS/TS/TSX/MJS, JSON, YAML, Markdown, CSS, GraphQL, TOML.
• tsp format owns: .tsp.
• Prettier (residual) owns only: .astro + .sh (slim config, astro+sh plugins only).

Editor / CI follow-ups (not yet applied)

• VS Code: install the oxc extension (oxc.oxc-vscode, oxfmt via LSP); set per-language editor.defaultFormatter (oxfmt for JS/TS/JSON/CSS/MD/YAML, TypeSpec ext for .tsp, Prettier for .astro/shell). Current shared .vscode/settings.json hardcodes esbenp.prettier-vscode.
• CI (.github/workflows/consistency.yml): replace the prettier-plugin build + pnpm run format:check with the oxfmt/tsp/residual pipeline (oxfmt needs no build; tsp format still needs the compiler built).

Risks / open items

1. One-time 55-file reformat lands a slightly different (but stable) JS/TS style vs Prettier.
2. Loss of automatic unused-import removal (mitigated by oxlint no-unused-vars).
3. .astro unsupported today — track oxc Tier-4 progress to eventually drop the Prettier residual.
4. oxfmt is Beta (Feb 2026); pin the version and re-verify churn on upgrades.
5. tsp format requires the compiler to be built before running (same as today's prettier-plugin build step).

---

Findings from a hands-on PoC that wired oxfmt/tsp format/residual-Prettier scripts side-by-side with the existing Prettier scripts (no repo-wide reformat performed).

[github-actions]

🎯 Issue Triage

Proposal to replace Prettier with oxfmt (oxc formatter) as the primary formatter for the monorepo, based on a hands-on PoC showing ~30× speed improvement (from ~80–100s down to ~9s combined pipeline) with minimal formatting churn (1.9% of files).

Category: Feature Request

Area: eng

🔍 Analysis

• Motivation: Prettier's format check takes 80–100s; the proposed oxfmt + tsp format + residual Prettier pipeline brings this to ~9s.
• Churn: 55/2947 JS/TS/MD files (1.9%) differ — primarily union-type line-breaking and arrow-arg wrapping style differences.
• Gaps: .astro (47 files) and .sh (1 file) are unsupported by oxfmt and would require residual Prettier. .tsp continues to use tsp format.
• Import organizing: prettier-plugin-organize-imports (sort + remove unused) cannot be directly replaced; sortImports in oxfmt uses a different algorithm. Decision needed: drop auto-remove or keep as residual step.
• Maturity risk: oxfmt is Beta (Feb 2026); version pinning and re-verification on upgrades needed.
• CI impact: consistency.yml would need updating to replace the Prettier pipeline.

📝 Suggested next steps

• Decide on import-organizing strategy (drop auto-removal or keep prettier-plugin-organize-imports as residual)
• Land one-time 55-file reformat PR if proceeding
• Update .github/workflows/consistency.yml to use the new pipeline
• Update .vscode/settings.json to set per-language default formatters
• Track oxc Tier-4 progress for .astro support to eventually drop the Prettier residual

Generated by Agentic Triage