Ship codex-exec-server in CLI and app-server packages (#57590)

## What changed

- Build and bundle the standalone executor in `bin/` for both package variants, and advertise its path through `execServer` in `codex-package.json`.
- Include the executor in Cargo and Bazel release packaging, macOS signing and notarization, and Windows signing verification. Preserve prebuilt signed binaries during packaging.
- Make installers detect a missing advertised executor while continuing to accept older packages without it.

## Testing

Add coverage for executor archive contents and executable permissions, signed binary preservation, and macOS entitlement verification. Add package smoke tests for initialization, version reporting, filesystem and process sandbox enforcement, and startup error parity with `codex exec-server`.

GitOrigin-RevId: 5c0027fc19f16cf9c2a5bb52dbd065a9776bc2af
This commit is contained in:
Dmitriy Kovalenko
2026-10-27 23:14:18 +00:00
committed by copyberry
parent a9ea49e78b
commit cf2c04b325
23 changed files with 497 additions and 55 deletions
@@ -208,7 +208,12 @@ def code_mode_host_debug_symbols(
tmp_path_factory: pytest.TempPathFactory,
) -> Path:
target = pytestconfig.getoption("package_target")
binaries = {"codex", "codex-app-server", "codex-code-mode-host"}
binaries = {
"codex",
"codex-app-server",
"codex-code-mode-host",
"codex-exec-server",
}
if "windows" in target:
binaries.update({"codex-command-runner", "codex-windows-sandbox-setup"})
@@ -4,7 +4,8 @@ ROOT_OF_EXTRACTED_PACKAGE
├── bin
│ ├── codex[.exe] # CLI package only
│ ├── codex-app-server[.exe] # app-server package only
│ └── codex-code-mode-host[.exe]
│ ├── codex-code-mode-host[.exe]
│ └── codex-exec-server[.exe]
├── codex-package.json
├── codex-path
│ └── rg[.exe]
@@ -18,11 +19,13 @@ Each package contains one entrypoint, not both codex and codex-app-server.
"""
import json
import os
import re
import shutil
import subprocess
import sys
import tomllib
from concurrent.futures import ThreadPoolExecutor
from pathlib import Path
import pytest
@@ -67,6 +70,261 @@ def test_cli_version_matches_release_sources(package: SmokePackage) -> None:
assert package.run("--version").stdout == f"codex-cli {version}\n"
@pytest.mark.parametrize("variant", ["cli_root", "app_server_root"])
def test_packaged_executor_initializes_and_enforces_sandbox(
package: SmokePackage, variant: str
) -> None:
"""Each package re-enters its executor helpers and enforces filesystem/process policy."""
root = getattr(package, variant)
windows = "windows" in package.target
suffix = ".exe" if windows else ""
executor = root / "bin" / f"codex-exec-server{suffix}"
directory = package.directory / f"{variant}-executor-sandbox"
directory.mkdir()
workspace = directory / "workspace"
workspace.mkdir()
home = directory / "home"
home.mkdir()
blocked = directory / "blocked.txt"
blocked.write_text("unchanged")
environment = {**package.environment, "CODEX_HOME": str(home)}
if windows:
# Elevated mode must discover the setup/runner in this package's resources.
# Exclude build outputs and host Codex installations from helper fallback.
environment["PATH"] = str(Path(os.environ["SystemRoot"]) / "System32")
sandbox = {
"permissions": {
"type": "managed",
"file_system": {
"type": "restricted",
"entries": [
{
"path": {"type": "special", "value": {"kind": "root"}},
"access": "read",
},
{
"path": {"type": "path", "path": workspace.as_uri()},
"access": "write",
},
],
},
"network": "restricted",
},
"cwd": workspace.as_uri(),
"windowsSandboxLevel": "elevated" if windows else "disabled",
}
version = (
subprocess.check_output(
[str(executor), "--version"], env=package.environment, text=True, timeout=45
)
.strip()
.split()[-1]
)
assert version == package.run("--version").stdout.strip().split()[-1]
with (
ThreadPoolExecutor(max_workers=1) as reader,
subprocess.Popen(
[str(executor), "--listen", "stdio"],
cwd=package.directory,
env=environment,
stdin=subprocess.PIPE,
stdout=subprocess.PIPE,
text=True,
) as process,
):
assert process.stdin is not None and process.stdout is not None
try:
process.stdin.write(
json.dumps(
{
"id": 1,
"method": "initialize",
"params": {"clientName": "package-smoke"},
}
)
+ "\n"
)
process.stdin.flush()
initialized = json.loads(
reader.submit(process.stdout.readline).result(timeout=45)
)
assert initialized["id"] == 1
assert initialized["result"]["sessionId"]
reported_environment = initialized["result"]["environmentInfo"]
metadata = json.loads((root / "codex-package.json").read_text())
assert reported_environment["executorVersion"] == metadata["version"]
process.stdin.write('{"method":"initialized","params":{}}\n')
# Restricted writes always launch the filesystem helper, even when allowed.
for request_id, path in [(2, workspace / "allowed.txt"), (3, blocked)]:
process.stdin.write(
json.dumps(
{
"id": request_id,
"method": "fs/writeFile",
"params": {
"path": path.as_uri(),
"dataBase64": "b2s=",
"sandbox": sandbox,
},
}
)
+ "\n"
)
process.stdin.flush()
# The first Windows helper can provision sandbox accounts before replying.
response = json.loads(
reader.submit(process.stdout.readline).result(
timeout=180 if windows and request_id == 2 else 45
)
)
if request_id == 2:
assert response == {"id": 2, "result": {}}, response
assert path.read_text() == "ok"
else:
assert response["id"] == 3 and "error" in response, response
assert blocked.read_text() == "unchanged"
command = (
[
str(Path(os.environ["SystemRoot"]) / "System32" / "cmd.exe"),
"/D",
"/S",
"/C",
r"echo process>process.txt & echo forbidden>..\blocked.txt & exit /b 0",
]
if windows
else [
"/bin/sh",
"-c",
"printf process >process.txt; printf forbidden >../blocked.txt; exit 0",
]
)
process.stdin.write(
json.dumps(
{
"id": 4,
"method": "process/start",
"params": {
"processId": "package-sandbox",
"argv": command,
"cwd": workspace.as_uri(),
"env": {
key: value
for key, value in environment.items()
if key.casefold() in {"path", "systemroot"}
},
"sandbox": sandbox,
"tty": False,
"pipeStdin": False,
"arg0": None,
},
}
)
+ "\n"
)
process.stdin.flush()
started = False
exit_code = None
notifications = []
while not started or exit_code is None:
response = json.loads(
reader.submit(process.stdout.readline).result(timeout=45)
)
notifications.append(response)
if response.get("id") == 4:
expected_type = (
"windowsRestrictedToken"
if windows
else "macosSeatbelt"
if "apple-darwin" in package.target
else "linuxSeccomp"
)
assert response["result"]["sandboxType"] == expected_type, response
started = True
if response.get("method") == "process/exited":
exit_code = response["params"]["exitCode"]
assert exit_code == 0, notifications
assert (workspace / "process.txt").is_file(), notifications
assert (workspace / "process.txt").read_text().strip() == "process"
assert blocked.read_text() == "unchanged"
process.stdin.close()
assert process.wait(timeout=45) == 0
finally:
if process.poll() is None:
process.kill()
def test_packaged_executor_matches_subcommand_startup(
package: SmokePackage,
tmp_path: Path,
) -> None:
"""Both binaries reject the same invalid executor arguments and config."""
home = tmp_path / "home"
temporary = tmp_path / "tmp"
home.mkdir()
temporary.mkdir()
(home / "config.toml").write_text("unknown_executor_option = true\n")
# Release builds require the helper directory to live outside their temp root.
environment = {
**package.environment,
"CODEX_HOME": str(home),
"TMPDIR": str(temporary),
"TEMP": str(temporary),
"TMP": str(temporary),
}
suffix = ".exe" if "windows" in package.target else ""
executor = package.cli_root / "bin" / f"codex-exec-server{suffix}"
for arguments in [
["--strict-config", "--listen", "stdio"],
["-c", "unknown_executor_option=true", "--strict-config", "--listen", "stdio"],
["--listen", "http://127.0.0.1:0"],
["--concurrent-requests", "0"],
[
"--aws-sigv4",
"--remote",
"https://example.openai.com",
"--environment-id",
"env-1",
],
["--remote-transport", "direct"],
["forward", "--connect", "ws://127.0.0.1:0"],
[
"-c",
"unknown_executor_option=true",
"forward",
"-c",
"analytics.enabled=false",
"--strict-config",
"--connect",
"ws://127.0.0.1:0",
"--remote",
"https://example.openai.com",
"--environment-id",
"env-1",
],
]:
outputs = []
for command in ([str(package.cli), "exec-server"], [str(executor)]):
result = subprocess.run(
[*command, *arguments],
cwd=tmp_path,
env=environment,
capture_output=True,
text=True,
timeout=45,
)
assert result.returncode != 0, arguments
outputs.append(
(
result.returncode,
result.stdout,
result.stderr.replace("codex exec-server", "codex-exec-server"),
)
)
assert outputs[0] == outputs[1], arguments
@pytest.mark.parametrize("entrypoint", ["codex", "codex-app-server"])
def test_app_server_runs_code_mode_through_python_sdk(
package: SmokePackage,