Files
codex/scripts/codex_package/README.md
Channing Conger ef500f4d23 Move code mode behind an IPC host
Split the code-mode protocol and client from the V8-backed runtime so core and
app-server no longer link codex-code-mode in production. ThreadManager now
provisions durable code-mode sessions through a shared external host process,
while tests can still inject the in-process provider.

The IPC protocol uses a persistent stdin/stdout transport. Each frame is a
4-byte big-endian length followed by JSON, with a 16 MiB frame limit. Client
requests carry u64 request IDs so create, execute, wait, terminate, and shutdown
operations can be multiplexed over one process. Session IDs isolate durable
stored values. Execute returns an ExecutionStarted response immediately and an
asynchronous InitialResponse when the initial yield or completion is available.

Nested tool calls and notifications travel from the host back to the client as
delegate requests with their own IDs. Delegate responses, cancellation, and
cell-closed lifecycle messages use the same framed channel. Wire operations
encode errors as Result values. A dead connection fails pending operations,
cancels outstanding delegates, and lets the provider spawn a new host for later
sessions.

Build codex-code-mode-host with V8 pointer-compression sandbox support and add
it to canonical primary and app-server packages, legacy Linux and Windows
bundles, signing verification, installers, Python runtime packages, and release
CI for macOS, Linux, and Windows. The host is discovered next to the current
executable, through CODEX_CODE_MODE_HOST_PATH, or on PATH. OS-level seccomp or
seatbelt restrictions remain a follow-up to this cross-platform process split.

Benchmarks were run from release builds on Linux x86_64 with the V8 sandbox
profile and a text('ok') workload. Cold measurements used 30 samples, warm
session provisioning used 200, and warm command execution used 500. Values are
mean/p50/p95 in milliseconds:

- session startup: in-process 0.002/0.002/0.005, IPC 2.623/2.599/2.894
- fresh-session command: in-process 1.831/1.758/1.915, IPC 7.428/7.252/8.306
- warm session provisioning: in-process 0.002/0.002/0.003,
  IPC 0.471/0.463/0.581
- warm command: in-process 1.759/1.757/1.940, IPC 2.005/2.001/2.166

The steady-state median command overhead is approximately 0.244 ms. The median
fresh host plus first command cost is 7.252 ms.

Validation:

- 62/62 core code-mode integration tests passed against the external host
- focused protocol, client, runtime, host, tools, and trace tests passed
- Cargo and Bazel real-process host IPC tests passed
- 11/11 package builder tests passed
- Bazel lock verification, scoped Clippy fixes, and repository formatting passed
2026-06-10 14:38:41 -07:00

79 lines
3.8 KiB
Markdown

# Codex package builder
This package contains the implementation behind `scripts/build_codex_package.py`.
The top-level script is the stable executable entry point; these modules keep the
package-building logic split by responsibility.
The builder creates a canonical Codex package directory:
```text
.
├── codex-package.json
├── bin
│ ├── <entrypoint>[.exe]
│ └── codex-code-mode-host[.exe]
├── codex-resources
│ ├── bwrap # Linux only
│ ├── zsh/bin/zsh # supported Unix targets only
│ ├── codex-command-runner.exe # Windows only
│ └── codex-windows-sandbox-setup.exe # Windows only
└── codex-path
└── rg[.exe]
```
The package directory is the primary artifact. Archive formats such as
`.tar.gz`, `.tar.zst`, and `.zip` are serializations of that directory.
If `--target` is omitted, the builder uses the release target for the current
host platform. On Linux, that default is a musl target to match Codex release
artifacts; pass a GNU Linux target explicitly for native glibc local builds. If
`--package-dir` is omitted, the builder creates a new temporary directory and
prints its path after the package is built.
The `--variant` flag selects the package entrypoint. Supported variants are
`codex` and `codex-app-server`. The `version` field in `codex-package.json` is
read from `[workspace.package].version` in `codex-rs/Cargo.toml`.
## Source-built artifacts
Artifacts built from this repository are built by the package builder in one
grouped `cargo build` command per package when they are needed and no prebuilt
override was provided:
- all targets: the selected entrypoint, unless `--entrypoint-bin` is provided
- all targets: `codex-code-mode-host`, unless `--code-mode-host-bin` is provided
- Linux targets: `bwrap`, unless `--bwrap-bin` is provided
- Windows targets: `codex-command-runner` and `codex-windows-sandbox-setup`,
unless the corresponding prebuilt helper flags are provided
The default cargo profile is `dev-small` because local iteration should favor
fast, small builds. Release jobs should pass `--cargo-profile release` and an
explicit target. Release jobs that already built and signed/notarized the
entrypoint should pass `--entrypoint-bin` so the package contains that exact
binary instead of rebuilding it.
Release jobs that already built package resource binaries should also pass the
corresponding resource flags: `--bwrap-bin` for Linux packages, and
`--codex-command-runner-bin` plus `--codex-windows-sandbox-setup-bin` for
Windows packages. This keeps package archive creation as a pure staging step
after signing instead of rebuilding resources.
When the builder source-builds an entrypoint for a Darwin or Linux target, it
downloads and verifies the matching Codex-built V8 release pair before invoking
Cargo and sets `RUSTY_V8_ARCHIVE` plus `RUSTY_V8_SRC_BINDING_PATH` for that
build. Windows targets keep Cargo's release-build MSVC artifact path. Explicit
overrides remain authoritative when both variables are already set. Set
`V8_FROM_SOURCE=1` to leave the build with the `v8` crate source-build path.
`rg` is not built from this repository, so the builder fetches it from the
DotSlash manifest at `scripts/codex_package/rg`. Downloaded archives are cached
under `$TMPDIR/codex-package/<target>-rg` and are reused only after the recorded
size and SHA-256 digest have been verified. Pass `--rg-bin` to use a local
ripgrep executable instead.
The patched zsh fork used by `shell_zsh_fork` is fetched from the DotSlash
manifest at `scripts/codex_package/codex-zsh` when the selected target has a
matching prebuilt artifact. Downloaded archives are cached under
`$TMPDIR/codex-package/<target>-zsh` and installed at
`codex-resources/zsh/bin/zsh`.