fix: prefer .ps1 over .cmd shim on Windows by fengmk2 · Pull Request #345 · voidzero-dev/vite-task

fengmk2 · 2026-04-18T06:21:36Z

Summary

Spawning a node_modules/.bin/*.cmd shim on Windows forces the OS to start cmd.exe, which intercepts Ctrl+C with a "Terminate batch job (Y/N)?" prompt and leaves the terminal in a corrupt state (backspace prints ^H, Ctrl+C prints ^C, input becomes sluggish).

When which() resolves a .cmd shim under node_modules/.bin/ and the system has pwsh.exe or powershell.exe available, the planner substitutes the sibling .ps1 path so the task graph records the script that actually runs. At spawn time, any .ps1 program is invoked via powershell.exe -File, so Ctrl+C propagates cleanly without the intermediate cmd.exe hop. If no PowerShell host is available, the substitution is skipped — the original .cmd path is preserved so the task still runs (matching pre-PR behavior).

Architecture

The logic is split across two layers so task metadata and cache keys reflect the real script, while the PowerShell host hop stays a spawn-time concern. Both layers consult the same POWERSHELL_HOST cache so plan-time and spawn-time decisions can never disagree.

Plan layer — vite_task_plan::ps1_shim

powershell_host() -> Option<&'static Arc<AbsolutePath>> — cached via LazyLock, prefers pwsh.exe over powershell.exe, None when neither is on PATH (or not on Windows). Owned by the plan crate but exposed as a pub accessor so the spawn layer can reuse the same cache.
prefer_ps1_sibling(AbsolutePathBuf) -> AbsolutePathBuf runs inside which() right after which::which_in resolves a binary. Gated on powershell_host().is_some(): if no host is available, the .cmd is kept untouched so we don't hand CreateProcess a .ps1 it can't execute.
On Windows with host + sibling .ps1: substitutes node_modules/.bin/vite.cmd with node_modules/.bin/vite.ps1. SpawnCommand.program_path and SpawnFingerprint reference the .ps1 directly; cache keys stay portable (workspace-relative).
.cmd/.bin/node_modules comparisons are case-insensitive.
No-op on non-Windows (compile-time const fn passthrough).

Spawn layer — vite_task::session::execute::powershell

wrap_ps1_with_powershell(&program, &args) -> Option<(program, args)>.
If the program ends in .ps1, wraps the spawn as powershell.exe -NoProfile -NoLogo -ExecutionPolicy Bypass -File <ps1> <args>, using the same host as the plan layer.
Option return lets the caller reuse the original Arcs on the passthrough path — no clones for non-.ps1 spawns.

Scope (what is not touched)

The plan-time substitution fires only when all of the following hold:

Extension is .cmd (case-insensitive), and
The file lives under node_modules/.bin/ (case-insensitive path-component check), and
A sibling .ps1 exists, and
pwsh.exe or powershell.exe is on PATH.

Only .cmd is considered because npm/pnpm/yarn's cmd-shim (the tool that generates the triplets) emits .cmd/.ps1/POSIX — never .bat.

Everything else is left alone:

.exe / .com / .bat binaries
System tools (C:\Windows\System32\where.cmd, chcp.com, etc.)
User-installed .cmd wrappers outside node_modules/.bin/
Any .cmd without a sibling .ps1
Stripped-down Windows installs with no PowerShell host on PATH (fall back to .cmd)

Closes voidzero-dev/vite-plus#1176

Test plan

cargo clippy --workspace --all-targets clean on macOS
cargo test --workspace passes on macOS — 4 ps1_shim tests + 2 powershell tests run cross-platform via injected host
Windows CI green
Manual verification on Windows: reproduce with https://github.com/Curtion/report-vite-plus-1 after bumping vite-task in vite-plus; Ctrl+C during vp run dev should leave the terminal clean
Manual verification: confirm pwsh.exe is preferred when both it and powershell.exe are on PATH
Manual verification: on a Windows box with PATH stripped of PowerShell, confirm .cmd is still used and tasks still run (regression guard for the Cursor-flagged "no host" case)

Address review feedback on #345: - Extract the 5 fixed PowerShell prefix flags to a `const POWERSHELL_PREFIX` - Build `new_args` via iterator chain instead of manual Vec pushes - Drop the no-op `let _ = (&program_path, &args);` on non-Windows

Pulls in voidzero-dev/vite-task#345 which prefers .ps1 shims over .cmd on Windows to avoid the "Terminate batch job (Y/N)?" prompt and terminal corruption on Ctrl+C during `vp run dev`. Closes #1176

Picks up the cache-portability fix for voidzero-dev/vite-task#345 (PowerShell rewrite moved from plan layer to spawn layer).

Picks up voidzero-dev/vite-task#345 fix for missing `which` dep on Windows after the module move.

fengmk2 · 2026-04-18T08:03:01Z

@cursor review

Pulls in voidzero-dev/vite-task#345 which prefers .ps1 shims over .cmd on Windows to avoid the "Terminate batch job (Y/N)?" prompt and terminal corruption on Ctrl+C during `vp run dev`. Closes #1176

Picks up the cache-portability fix for voidzero-dev/vite-task#345 (PowerShell rewrite moved from plan layer to spawn layer).

Picks up voidzero-dev/vite-task#345 fix for missing `which` dep on Windows after the module move.

cursor

✅ Bugbot reviewed your changes and found no new issues!

Comment @cursor review or bugbot run to trigger another review on this PR

^{Reviewed by Cursor Bugbot for commit 4f14ebe. Configure here.}

fengmk2 · 2026-04-18T13:15:12Z

before

after

Spawning a .cmd shim (e.g. `vite.cmd`) from any shell forces Windows to start `cmd.exe`, which intercepts Ctrl+C with a "Terminate batch job (Y/N)?" prompt and leaves the terminal in a corrupt state. When the resolved binary is `.cmd`/`.bat` and a sibling `.ps1` exists, route the spawn through `powershell.exe -File` (preferring `pwsh.exe`) so Ctrl+C propagates cleanly without the cmd.exe hop. Closes voidzero-dev/vite-plus#1176

Address review feedback on #345: - Extract the 5 fixed PowerShell prefix flags to a `const POWERSHELL_PREFIX` - Build `new_args` via iterator chain instead of manual Vec pushes - Drop the no-op `let _ = (&program_path, &args);` on non-Windows

Cursor review flagged that applying the rewrite at plan time leaks absolute `.ps1` paths and `powershell.exe` into `SpawnFingerprint`, breaking cache key portability across machines, CI, and repo moves (the cache was designed around relative paths). Move the rewrite to the spawn layer (`session/execute/spawn.rs`) so the fingerprint still keys on the original tool's relative path and args, while the actual spawn still routes through PowerShell.

The `.cmd`→`.ps1` rewrite moved from `vite_task_plan` (where `which` was already a dep) to `vite_task` without bringing the dep along. Windows CI failed with E0433. Only needed at `cfg(windows)`.

- Return Option<(Arc, Arc)>; caller reuses original references on the passthrough path, dropping two unconditional Arc clones per spawn - Inline the `imp` submodule: two cfg-gated definitions replace the `cfg_attr(expect(missing_const_for_fn))` workaround - Use eq_ignore_ascii_case to drop the lowercase String allocation - Gate pure rewrite on `any(windows, test)` so tests run on every OS

System or user-installed .cmd/.bat files elsewhere on PATH (e.g., `C:\Windows\System32\where.cmd`) must not be silently rerouted through PowerShell — the triplet shape (.cmd + .ps1 + POSIX) is only guaranteed for npm/pnpm/yarn shims. Check the last two path components before rewriting. Adds a negative test for a .cmd outside node_modules/.bin.

- Compare `.bin` / `node_modules` path components case-insensitively, matching the existing case-insensitive `.cmd`/`.bat` check and npm's de-facto lowercase naming (robust to symlinks/mounts) - Condense the scope-check comment to one line to match the repo style - Narrow `#[expect(clippy::disallowed_types)]` from the whole test module to just the two helpers that actually bridge std paths

fengmk2 · 2026-04-18T14:25:58Z

@cursor review

cursor

✅ Bugbot reviewed your changes and found no new issues!

Comment @cursor review or bugbot run to trigger another review on this PR

^{Reviewed by Cursor Bugbot for commit 3e1fc38. Configure here.}

…t spawn Record the script that actually runs in task metadata instead of the `.cmd` shim that never spawns: - New `vite_task_plan::ps1_shim::prefer_ps1_sibling` substitutes `node_modules/.bin/*.cmd` (or `.bat`) with the sibling `.ps1` when resolving binaries via `which()`. Applied case-insensitively and only under `node_modules/.bin` so system tools stay untouched. - `SpawnCommand.program_path` / `SpawnFingerprint` now reference the `.ps1` directly; cache keys stay portable because the path is still relative to the workspace. - `vite_task::session::execute::powershell` shrinks to just the powershell wrapping: if a program ends in `.ps1`, wrap the spawn as `powershell.exe -NoProfile -NoLogo -ExecutionPolicy Bypass -File <ps1> <args>`. No more filesystem stat or path-shape check at spawn time. - Tests split accordingly: sibling selection is covered in ps1_shim; spawn tests only verify the powershell wrap.

- Swap `cfg_attr(not(windows), expect(missing_const_for_fn))` for two cfg-gated function definitions, matching the pattern already used in `session::execute::powershell` - Move `#[expect(clippy::disallowed_types)]` from the test module down to the two helper functions that actually bridge tempdir's std paths

fengmk2 · 2026-04-18T14:48:23Z

@cursor review

cursor

✅ Bugbot reviewed your changes and found no new issues!

Comment @cursor review or bugbot run to trigger another review on this PR

^{Reviewed by Cursor Bugbot for commit 26021ca. Configure here.}

npm/pnpm/yarn's cmd-shim only emits `.cmd`/`.ps1`/POSIX triplets — `.bat` never shows up in a `node_modules/.bin` shim. The extra check was noise.

fengmk2 · 2026-04-19T07:32:51Z

@cursor review

Cursor flagged: `prefer_ps1_sibling` was substituting `.cmd` → `.ps1` unconditionally at plan time. If neither `pwsh.exe` nor `powershell.exe` is on PATH, spawn-time `wrap_ps1_with_powershell` returns `None`, and `CreateProcess` then tries to run the raw `.ps1` and fails. Before this PR the `.cmd` would have worked (with the Ctrl+C bug, but at least it ran). Gate the plan-time substitution on PowerShell host availability: - Move `POWERSHELL_HOST` from spawn layer into `ps1_shim` and expose it via `pub fn powershell_host()`. One cached lookup shared by both layers — plan decides only when spawn will actually be able to invoke the `.ps1`. - `prefer_ps1_sibling` returns the input unchanged when `powershell_host()` is `None`, so PowerShell-less Windows systems keep using the `.cmd` path (same behavior as before this PR). - `session::execute::powershell` reuses the shared host, dropping its local `LazyLock` and the now-unused `which` dep on `vite_task`.

fengmk2 · 2026-04-19T07:50:04Z

@cursor review

cursor

✅ Bugbot reviewed your changes and found no new issues!

Comment @cursor review or bugbot run to trigger another review on this PR

^{Reviewed by Cursor Bugbot for commit be08530. Configure here.}

fengmk2 self-assigned this Apr 18, 2026

fengmk2 mentioned this pull request Apr 18, 2026

chore: bump vite-task to e0229ffb voidzero-dev/vite-plus#1414

Draft

This comment was marked as resolved.

Sign in to view

cursor bot reviewed Apr 18, 2026

View reviewed changes

Comment thread crates/vite_task_plan/src/plan.rs Outdated

This comment was marked as resolved.

Sign in to view

fengmk2 added a commit to voidzero-dev/vite-plus that referenced this pull request Apr 18, 2026

chore: bump vite-task to c7523684

ceaac3f

Picks up the cache-portability fix for voidzero-dev/vite-task#345 (PowerShell rewrite moved from plan layer to spawn layer).

fengmk2 added a commit to voidzero-dev/vite-plus that referenced this pull request Apr 18, 2026

chore: bump vite-task to 7b74d616

8aa7edc

Picks up voidzero-dev/vite-task#345 fix for missing `which` dep on Windows after the module move.

This comment was marked as resolved.

Sign in to view

fengmk2 added a commit to voidzero-dev/vite-plus that referenced this pull request Apr 18, 2026

chore: bump vite-task to c7523684

bbe904f

Picks up the cache-portability fix for voidzero-dev/vite-task#345 (PowerShell rewrite moved from plan layer to spawn layer).

fengmk2 added a commit to voidzero-dev/vite-plus that referenced this pull request Apr 18, 2026

chore: bump vite-task to 7b74d616

249042f

Picks up voidzero-dev/vite-task#345 fix for missing `which` dep on Windows after the module move.

cursor bot reviewed Apr 18, 2026

View reviewed changes

fengmk2 changed the title ~~fix(plan): prefer .ps1 over .cmd shim on Windows~~ fix(spawn): prefer .ps1 over .cmd shim on Windows Apr 18, 2026

fengmk2 added 8 commits April 18, 2026 22:24

fix(task): add missing which dep on Windows

c2738fa

The `.cmd`→`.ps1` rewrite moved from `vite_task_plan` (where `which` was already a dep) to `vite_task` without bringing the dep along. Windows CI failed with E0433. Only needed at `cfg(windows)`.

chore: update Cargo.lock for vite_task which dep

89ace99

fengmk2 force-pushed the fix/prefer-ps1-over-cmd-shim branch from 5ce6c62 to 3e1fc38 Compare April 18, 2026 14:25

cursor bot reviewed Apr 18, 2026

View reviewed changes

fengmk2 changed the title ~~fix(spawn): prefer .ps1 over .cmd shim on Windows~~ fix: route node_modules/.bin .cmd shims through PowerShell on Windows Apr 18, 2026

cursor bot reviewed Apr 18, 2026

View reviewed changes

fengmk2 changed the title ~~fix: route node_modules/.bin .cmd shims through PowerShell on Windows~~ refactor: prefer .ps1 over .cmd shim on Windows Apr 18, 2026

fengmk2 changed the title ~~refactor: prefer .ps1 over .cmd shim on Windows~~ fix: prefer .ps1 over .cmd shim on Windows Apr 18, 2026

fengmk2 commented Apr 19, 2026

View reviewed changes

Comment thread crates/vite_task_plan/src/ps1_shim.rs Outdated

refactor(ps1_shim): drop .bat handling

bed8407

npm/pnpm/yarn's cmd-shim only emits `.cmd`/`.ps1`/POSIX triplets — `.bat` never shows up in a `node_modules/.bin` shim. The extra check was noise.

fengmk2 marked this pull request as ready for review April 19, 2026 07:32

fengmk2 requested a review from branchseer April 19, 2026 07:33

cursor bot reviewed Apr 19, 2026

View reviewed changes

Comment thread crates/vite_task_plan/src/ps1_shim.rs

fengmk2 added 2 commits April 19, 2026 15:45

refactor(spawn): use cfg-gated import for powershell_host

be08530

cursor bot reviewed Apr 19, 2026

View reviewed changes

Conversation

fengmk2 commented Apr 18, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Summary

Architecture

Scope (what is not touched)

Test plan

Uh oh!

This comment was marked as resolved.

Uh oh!

This comment was marked as resolved.

This comment was marked as resolved.

Uh oh!

fengmk2 commented Apr 18, 2026

Uh oh!

cursor bot left a comment

Choose a reason for hiding this comment

Uh oh!

fengmk2 commented Apr 18, 2026

Uh oh!

fengmk2 commented Apr 18, 2026

Uh oh!

cursor bot left a comment

Choose a reason for hiding this comment

Uh oh!

fengmk2 commented Apr 18, 2026

Uh oh!

cursor bot left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

fengmk2 commented Apr 19, 2026

Uh oh!

Uh oh!

fengmk2 commented Apr 19, 2026

Uh oh!

cursor bot left a comment

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

fengmk2 commented Apr 18, 2026 •

edited

Loading