Skip to content

[Bug]: Windows: Antigravity install rolls back after successful validation because killing the already-exited validation process fails #32457

Open

Description

@Ufotor06

Before submitting

  • I searched existing issues and did not find a duplicate.
  • I included enough detail to reproduce or investigate the problem.

Related but different: #29273 / #29279 (the rename after validation fails with EPERM). Here the install never reaches the rename: it fails in the validation scope's finalizer.

Area

apps/server

Steps to reproduce

  1. Windows 11 x64, fresh T3 Code (Beta) 0.1.7 desktop install.
  2. Settings → Providers → Antigravity → enable → Install.
  3. Wait for download, extraction and verification.

Reproduced twice in a row.

Expected behavior

Validation succeeds, so the runtime is published to versions/<sha256> and activated.

Actual behavior

  • UI: "Could not finish the Antigravity installation. Check disk space and directory access." The provider then shows "Not found · Antigravity ACP runtime is not installed."
  • Disk space is fine (358 GB free) and Windows Defender logged nothing.
  • In server.trace.ndjson, AntigravityInstallation.validate ends with Success (~26 s). AcpSessionRuntime.recordTermination follows. Then AntigravityInstallation.install fails with PlatformError: Unknown: ChildProcess.kill (...agy_acp_server.exe) / Error: Failed to kill child process, and the staging dir is removed.

On win32, validate in apps/server/src/provider/AntigravityInstallation.ts (still on main, ~line 395) registers:

yield* Scope.addFinalizer(
  validationScope,
  child.kill({ forceKillAfter: "1 second" }).pipe(Effect.orDie),
);

By the time the finalizer runs, the process tree has already exited, so the kill fails. orDie turns that into a defect, which fails the whole install even though validation passed. A kill failure on a process that has already exited should probably be ignored or just logged, not treated as fatal. The fallback message also hides the real cause; #31050 handles a similar case for signals and exit codes.

Workaround: download the same pinned zip manually (SHA-256 matches), extract both exes into one folder, and set Binary path. The provider then becomes ready, signs in and loads models normally.

Impact

Blocks work completely

Version or commit

T3 Code (Beta) 0.1.7 desktop (server 0.1.7). Same code on main at d14f87f.

Environment

Windows 11 Home 10.0.26200, win32-x64, agy-acp-server 1.4.5 (windows-x86_64)

Logs or stack traces

AntigravityInstallation.validate  26186ms  Success
AntigravityInstallation.install  174672ms  Failure
PlatformError: Unknown: ChildProcess.kill (C:\Users\<user>\.t3\tools\antigravity-acp\win32-x64\versions\.install-Mb5aKw\runtime\agy_acp_server.exe)
    at systemError (.../ManagedEndpointRuntime-olGvALiB.mjs:4656:34)
    at toPlatformError (.../ManagedEndpointRuntime-olGvALiB.mjs:122506:53)
    at .../ManagedEndpointRuntime-olGvALiB.mjs:122780:18
    at AntigravityInstallation.install (.../runtimeLayer-BDHHK7u1.mjs:49113:16)
    at ProviderInstallation.start (.../server-UENsecr6.mjs:25080:70)
    at ws.rpc.provider.install.start (.../server-UENsecr6.mjs:19898:23) {
  [cause]: Error: Failed to kill child process
      at .../ManagedEndpointRuntime-olGvALiB.mjs:122779:18
}

Activity

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions