denoland/deno · error · anyhow::Error

failed to run native tsc: {e}

Error message

failed to run native tsc: {e}

What it means

After obtaining the native compiler, deno spawns it with cwd set to the project root and PWD pinned to the same directory. If the OS-level spawn itself fails — executable missing, not executable, or a fork/exec error — the io error is wrapped in this message.

Source

Thrown at cli/tsc/native.rs:234

    .arg("--project")
    .arg(tsconfig_path)
    .arg("--noEmit")
    .arg("--pretty")
    .arg("false")
    .arg("--diagnostics")
    .current_dir(project_root)
    // The native compiler is written in Go, whose `os.Getwd` trusts the `PWD`
    // environment variable over `getcwd()`. `current_dir` calls `chdir` but
    // does not update the inherited `PWD`, so a symlinked launch directory
    // (e.g. `/tmp` -> `/private/tmp` on macOS) would leave `PWD` pointing at
    // the symlink and make tsc report every path relative to it. Pin `PWD` to
    // the same directory we chdir'd into.
    .env("PWD", project_root)
    .stdout(Stdio::piped())
    .stderr(Stdio::piped())
    .output()
    .await
    .map_err(|e| anyhow::anyhow!("failed to run native tsc: {e}"))
}

/// A single `path(line,col): error TS####: message` line from `tsc --pretty
/// false`. Continuation lines (indented elaborations) are folded into the
/// preceding diagnostic's message.
static DIAGNOSTIC_RE: LazyLock<Regex> = LazyLock::new(|| {
  Regex::new(
    r"^(?P<file>.+?)\((?P<line>\d+),(?P<col>\d+)\): (?P<cat>error|warning|message) TS(?P<code>\d+): (?P<msg>.*)$",
  )
  .unwrap()
});

/// The position-less form (`error TS####: message`), e.g. `TS18003` (no
/// inputs) or a `TS5###` config error.
static DIAGNOSTIC_NO_POS_RE: LazyLock<Regex> = LazyLock::new(|| {
  Regex::new(r"^(?P<cat>error|warning|message) TS(?P<code>\d+): (?P<msg>.*)$")
    .unwrap()
});

View on GitHub (pinned to 89f33cbef2)

Solutions

1. Delete the cached compiler directory ($DENO_DIR/tsc/<version>/<platform>) so the next run re-downloads and re-extracts it with correct permissions
2. Check the binary is executable and DENO_DIR is not mounted noexec: `ls -l <tsc-path>` and inspect mount options
3. Set DENO_TSC_BIN to a known-good tsc binary to bypass the cached copy

Example fix

# before: cached copy lost its exec bit
ls -l "$DENO_DIR/tsc/7.0.2/linux-x64/lib/tsc"   # missing x
# after
rm -rf "$DENO_DIR/tsc/7.0.2"
deno check mod.ts   # triggers a clean re-download

# bash: verify the cached binary is executable before running deno
tscbin="$DENO_DIR/tsc/7.0.2/$(case "$(uname -s)" in Linux) echo linux;; Darwin) echo darwin;; *) echo win32;; esac)-$(uname -m | sed 's/x86_64/x64/;s/aarch64/arm64/')/lib/tsc"
if [ -e "$tscbin" ] && [ ! -x "$tscbin" ]; then rm -rf "$(dirname "$(dirname "$tscbin")")"; fi

# bash: clear the cache and retry once on spawn failure
if ! deno check mod.ts 2>err.log; then
  grep -q "failed to run native tsc" err.log && rm -rf "$DENO_DIR/tsc" && exec deno check mod.ts
  exit 1
fi

When it happens

Trigger: The tsc path exists in the cache but is not executable (lost +x bit, extracted from a zip), the file vanished between the existence check and spawn, or process creation fails (memory limits, container restrictions).

Common situations: Cache corruption; extracting the toolchain without preserving permissions; DENO_DIR on a noexec mount; tightly limited containers.

Related errors

• The bench name can't be empty
• Expected 'fn' field in the first argument to be a bench func
• BenchContext::end() has already been invoked
• Object is not a valid image or a path to an image. `Deno.jup
• Request url protocol must be 'http:' or 'https:': received '

AI-assisted analysis of denoland/deno@89f33cbef2 (2026-08-16). Data as JSON: /api/errors/ded367599e7adf80. Report an issue: GitHub.