@smbcloud / cli
Code Commits Issues Pulls Cloud Agent CI/CD
development
claude/smbcloud-github-app-k9641p development feature/apps feature/build-release feature/ci-docs feature/ci-ruby-gems feature/cicd feature/completion feature/dist-nuget feature/fix-ci-sdk-pypi-rubygem feature/fix-clippy-on-stable-channel feature/fix-deployment-script feature/github-email feature/gresiq-v1 feature/project-new-structure feature/rsync-deploy feature/ruby-runner feature/sdk-npm feature/sdk-workflows feature/smbcloud-cli-release-npm feature/smbcloud-deploy feature/wasm main release/v0.7.4
md 3.62 KB
View source Raw

Plan: decouple build/deploy from the CLI into smbcloud-deploy

Status: in progress. Stages 0 to 3 (vite_spa slice) shipped. Remaining work below is the next milestone.

Goal

Move the deploy logic out of crates/cli into a reusable crates/smbcloud-deploy engine so the CLI, the CI action, and (later) the server-side git receiver all drive the same code. They differ only in how they report progress and how they authenticate.

Two inversions make it reusable: - Reporter replaces direct spinners / dialoguer / println!, so the engine never owns the terminal. The CLI supplies SpinnerReporter. - Auth is passed in (a token / credentials); the engine never reads ~/.smb or prompts for login.

Decisions already made: - Networking crates (smbcloud-networking*) stay direct deps of the engine. Only reporting and auth are inverted. - git.rs (SSH-git remote setup) stays in the CLI for now. It is the SSH-git path slated for replacement by git-smart-HTTP, so decoupling it is wasted work. - Interactive setup (setup*.rs, process_migrate.rs) stays in the CLI. It is UX, not engine.

Done

  • Stage 0 engine crate: error::DeployError, report::{Reporter, NoopReporter}, runner::detect_runner.
  • Stage 1 CLI depends on the engine; ui/reporter.rs::SpinnerReporter; detect_runner rewired; old deploy/detect_runner.rs removed.
  • Stage 2 transport: transport::{Transport, RsyncTransport} + moved known_hosts. CLI helper deploy::rsync_transport(config, runner, user_id) resolves the local ~/.ssh identity and remote path.
  • Stage 3 (vite_spa slice) build split: build::{BuildStrategy, BuildArtifact, ViteSpaBuild}. process_deploy_vite_spa.rs now calls the engine for the build and reuses one SpinnerReporter for build + transport.

Next milestone

1. Shared remote-command step

Several strategies do the same "SSH in and restart the service" step, inlined per file. Extract it into the engine first (e.g. transport::RemoteCommand or a remote module) so the strategies can share it. Model it behind the same pinned host-key SSH used by RsyncTransport.

2. Remaining BuildStrategy extractions (one per pass, compile-green each)

Do smallest first. Each: lift build mechanics into a BuildStrategy, swap spinners + stop_and_persist for Reporter, take auth as a param, route the remote-restart step through step 1.

File Lines Shape
process_deploy_rails.rs ~395 rsync shared lib, SSH compile native gem, git force-push
process_deploy_rust.rs ~641 cross-build Linux binary, rsync, SSH restart
process_deploy_nextjs_ssr.rs ~700 install, build, upload .next/standalone, SSH restart
process_deploy_swift.rs ~832 Docker Linux build, rsync binary + Resources, SSH restart

Follow ViteSpaBuild as the template. Keep the deployment-record API calls (create_deployment / update) in the CLI router, not the engine.

3. Stage 4: router + retire the CommandResult wart

  • Flip process_deploy.rs's router to call an engine::deploy(...) orchestrator.
  • Retire CommandResult's spinner: Spinner field (the "return a live spinner, caller stops it" pattern) now that the engine reports the whole flow through Reporter. CommandResult has ~50 construction sites, so this is its own pass: either make spinner optional or drop it and have commands return plain data.

Reference

  • Spec: smbcloud repo .agents/specs/smbcloud-deploy.md (product direction), smbcloud-actions.md (CI wrapper). Direction is git-smart-HTTP, local-first build, drop SSH transport later.
  • Engine crate: crates/smbcloud-deploy/.