tsrs-doctor

Compare tsgo (TypeScript 7) and tsrs, the Rust port of the TypeScript compiler, on your own repo: one command for a diagnostics diff, a speed and memory benchmark, and VS Code setup.

The npm package is not published yet. Run the CLI from a source checkout:

git clone https://github.com/majiayu000/tsrs-doctor.git
cd tsrs-doctor
bun install --frozen-lockfile
bun bin/tsrs-doctor.js /path/to/your-project

performance, median of 5 runs:
                wall                 cpu (user+sys)          peak RSS
  tsgo          7.17s (6.27s–7.47s)  22.57s (21.37s–24.16s)  7562 MiB (7212 MiB–7916 MiB)
  tsrs          2.29s (2.13s–2.61s)  8.40s (8.28s–8.75s)     1881 MiB (1877 MiB–1893 MiB)
  tsrs vs tsgo  3.1× faster                                  4.0× less memory

diagnostics: tsgo 0, tsrs 0; same 0, missing in tsrs 0, extra in tsrs 0, text differs 0

tsrs is 3.1× faster and uses 4.0× less memory than tsgo on this repo, identical diagnostics

Real output on type-fest (Apple Silicon). More in examples/.

How it compares

	tsrs-doctor	tsrs bench/run.py	ts-checker-bench	tsc-diagnostics-diff-action
Runs on your repo	✅	❌ fixed projects	❌ fixed projects	✅ one compiler
One command, no tsrs build	✅	❌ cargo build	⚠️ clone the repo	✅
Wall time, CPU, peak RSS	✅	✅	✅	❌
Diagnostics matched by file, line and code	✅	❌ counts only	⚠️ +n / −m	⚠️ text diff
Flags tsgo run-to-run noise and Go-mode-only differences	✅	❌	❌	❌
TypeScript version-skew warning	✅	⚠️ docs	⚠️ footnote	❌
VS Code setup for tsrs (#661)	✅	❌ manual symlink	❌	❌

The full feature comparison, including the tsrs companion projects, is in docs/comparison.md.

Why

The tsrs README asks users for feedback on performance, memory and conformance, but you can't get a useful answer from diffing two compiler outputs:

• tsgo's diagnostic order changes from run to run, so a plain diff is mostly noise.
• Some differences come from checker order or lazy members and go away with --checkerAssignment go --noLazyMembers. Others are real.
• Your project's tsgo nightly may not be the TypeScript commit that tsrs ports.
• Running the npm launchers adds Node startup time and measures Node's memory, not the compiler's.

tsrs-doctor handles all four.

What it does

1. Finds both compilers. It uses the project's own tsrs and typescript@7 / @typescript/native-preview when they are installed. Otherwise it falls back to the pair it bundles: tsrs 0.18.0 and the exact TypeScript 7.1 nightly that tsrs ports. It always runs the native binaries directly.

2. Times them fairly. Runs alternate between the two compilers (--runs, default 3), each in a fresh process with --noEmit and no incremental cache. It reports the median and range of wall time, CPU time and peak RSS.

3. Aligns diagnostics by file, line and error code, so output order doesn't matter. Each difference is classified as missing in tsrs, extra in tsrs, or same place with different text. Differences can also be tagged:

   • baseline unstable: tsgo itself didn't agree across runs.
   • gone in Go mode: tsrs re-run with --checkerAssignment go --noLazyMembers --checkers 4 no longer differs.

   An untagged difference is the kind worth reporting upstream.

4. Warns about version skew between the TypeScript commit tsrs ports and your tsgo.

5. Writes reports: --json for CI and --md to paste into an issue.

Exit codes: 0 identical diagnostics, 1 differences or a crash/timeout, 2 tool error.

Install

Requires Bun. Supported platforms are macOS arm64 and Linux x64/arm64, the ones tsrs ships binaries for.

Until the npm package is published, use the source checkout above. From any directory:

bun /absolute/path/to/tsrs-doctor/bin/tsrs-doctor.js /path/to/your-project

Usage

tsrs-doctor [dir] [options] [-- <compiler args>]
tsrs-doctor vscode [dir] [--undo]

  -p, --project <path>   tsconfig, relative to dir (default: tsconfig.json)
      --runs <n>         timed runs per compiler (default: 3)
      --tsrs <bin>       native tsrs binary (default: auto-detect)
      --baseline <bin>   baseline compiler, tsgo or tsc (default: auto-detect)
      --timeout <sec>    per-run timeout (default: 600)
      --json <file>      full report as JSON
      --md <file>        full report as Markdown
  -- <args>              passed to both compilers, e.g. -- --checkers 4

In a monorepo whose root tsconfig.json only lists references, tsrs-doctor stops and prints a -p command for each referenced project, because that config checks no files.

In CI

- uses: oven-sh/setup-bun@v2
- run: bun install
- uses: actions/checkout@v4
  with:
    repository: majiayu000/tsrs-doctor
    path: .tools/tsrs-doctor
- run: bun install --frozen-lockfile
  working-directory: .tools/tsrs-doctor
- run: bun .tools/tsrs-doctor/bin/tsrs-doctor.js . --md tsrs-report.md
- if: always()
  run: cat tsrs-report.md >> "$GITHUB_STEP_SUMMARY"

The step fails (exit 1) when tsrs and tsgo disagree. The Markdown report also includes a badge you can paste into your README.

VS Code

The TypeScript 7 extension looks for an executable named tsgo or tsc in js/ts.tsdk.path, and the tsrs package ships neither (oxc-project/tsrs#661). The documented workaround is a hand-made symlink in your home directory, redone on every machine.

bun add -d tsrs
bun /absolute/path/to/tsrs-doctor/bin/tsrs-doctor.js vscode .

This writes two things:

• .vscode/tsrs-tsdk/tsgo: a three-line sh script that runs node_modules/.bin/tsrs. It contains no absolute paths, so you can commit it and every teammate gets the tsrs from the lockfile.
• Two keys merged into .vscode/settings.json, keeping your comments and other settings: "js/ts.tsdk.path": ".vscode/tsrs-tsdk" and "js/ts.experimental.useTsgo": true.

Before it writes anything, it backs up the original settings into .vscode/tsrs-tsdk/ (git-ignored). Afterwards it checks that the shim really starts tsrs. bun /absolute/path/to/tsrs-doctor/bin/tsrs-doctor.js vscode . --undo puts back the previous values of those two keys, leaves any other edits you made, and removes the directory.

In VS Code: install TypeScriptTeam.native-preview, trust the workspace, accept the prompt to use the workspace version, then run "TypeScript 7: Restart Server".

Example results

project	config	tsgo	tsrs	speed	memory	diagnostics
type-fest	tsconfig.json	7.17s / 7562 MiB	2.29s / 1881 MiB	3.1×	4.0×	identical (0)
hono	tsconfig.spec.json	0.10s / 196 MiB	0.05s / 94 MiB	1.8×	2.1×	identical (6)
ky	tsconfig.json	0.06s / 67 MiB	0.03s / 42 MiB	1.9×	1.6×	identical (0)

Measured on an Apple Silicon Mac, 2026-10-22, against typescript@7.1.0-dev.20260930.4, the nightly tsrs 0.18.0 ports.

hono shows why the version check matters. Its own @typescript/native-preview@7.0.0-dev.20260210 disagrees with tsrs on 3 diagnostics. Against the matching 7.1 nightly the diagnostics are identical, so all 3 come from TypeScript changes between the two nightlies, not from tsrs. Two of them are a 7.1 regression (microsoft/TypeScript#64431, first reported against tsrs as oxc-project/tsrs#664) that tsrs faithfully ports.

License

MIT