go-scm/docs/development.md
Snider 4558504499 docs: graduate TODO/FINDINGS into production documentation
Replace internal task tracking (TODO.md, FINDINGS.md) with structured
documentation in docs/. Trim CLAUDE.md to agent instructions only.

Co-Authored-By: Virgil <virgil@lethean.io>
2026-02-20 15:02:22 +00:00

10 KiB

go-scm Development Guide


Prerequisites

  • Go 1.25 or later
  • Git (for git/ package integration tests)
  • gh CLI (for collect/github.go and rate limit checking — not required for unit tests)
  • SSH access to agent machines (for agentci/ integration — not required for unit tests)
  • Access to forge.lthn.ai/core/go for the framework dependency

Repository Layout

go-scm/
├── go.mod              Module definition (forge.lthn.ai/core/go-scm)
├── forge/              Forgejo API client + tests
├── gitea/              Gitea API client + tests
├── git/                Multi-repo git operations + tests
├── agentci/            Clotho Protocol + security + tests
├── jobrunner/          Poller, journal, types + tests
│   ├── forgejo/        Forgejo signal source + tests
│   └── handlers/       Pipeline handlers + tests
├── collect/            Data collection pipeline + tests
└── docs/               Architecture, development, history

Building

This module has no binary targets — it is a library. Build validation is via tests and go vet:

go build ./...         # Compile all packages
go vet ./...           # Static analysis

Testing

Run all tests

go test ./...

Run a specific test or package

go test -v -run TestName ./forge/
go test -v ./agentci/...

Run with race detector

go test -race ./...

Race detection is particularly important for git/ (parallel status), jobrunner/ (concurrent poller cycles), and collect/ (concurrent rate limiter).

Coverage

go test -coverprofile=cover.out ./...
go tool cover -html=cover.out

Current coverage targets (as of Phase 3 completion):

Package Coverage
forge/ 91.2%
gitea/ 89.2%
git/ 96.7%
agentci/ 94.5%
jobrunner/ 86.4%
jobrunner/forgejo/ 95.0%
jobrunner/handlers/ 83.8%
collect/ 83.0%

Local Dependencies

go-scm depends on forge.lthn.ai/core/go for the framework, logging, config, and IO abstractions. The go.mod file includes a replace directive pointing to a sibling directory:

replace forge.lthn.ai/core/go => ../go

Preferred approach: use a go.work file in your workspace root to avoid editing go.mod for local development:

// go.work
go 1.25

use (
    ./go
    ./go-scm
)

With a workspace file in place, the replace directive in go.mod is superseded and can be left as a fallback.


Test Patterns

forge/ and gitea/ — httptest mock server

Both SDK wrappers require a live HTTP server because the Forgejo SDK makes an HTTP GET to /api/v1/version during client construction. Use net/http/httptest for all tests:

func setupServer(t *testing.T) (*forge.Client, *httptest.Server) {
    mux := http.NewServeMux()
    mux.HandleFunc("/api/v1/version", func(w http.ResponseWriter, r *http.Request) {
        w.Header().Set("Content-Type", "application/json")
        json.NewEncoder(w).Encode(map[string]string{"version": "1.20.0"})
    })
    // ... register other handlers ...
    srv := httptest.NewServer(mux)
    t.Cleanup(srv.Close)
    client, err := forge.New(srv.URL, "test-token")
    require.NoError(t, err)
    return client, srv
}

SDK route patterns sometimes differ from the public API documentation. Notable divergences discovered during test construction:

  • CreateOrgRepo uses /api/v1/org/{name}/repos (singular org)
  • ListOrgRepos uses /api/v1/orgs/{name}/repos (plural orgs)

Config isolation: always isolate the config file from the real development machine during tests:

t.Setenv("HOME", t.TempDir())
t.Setenv("FORGE_TOKEN", "test-token")
t.Setenv("FORGE_URL", srv.URL)

Gitea mirror validation: CreateMirrorRepo with Service: GitServiceGithub requires a non-empty AuthToken. The SDK rejects the request locally before sending to the server if the token is absent.

git/ — real git repositories

git/ tests use real temporary git repositories rather than mocks. The standard setup pattern:

func setupRepo(t *testing.T) string {
    dir := t.TempDir()
    run := func(args ...string) {
        cmd := exec.Command("git", args...)
        cmd.Dir = dir
        require.NoError(t, cmd.Run())
    }
    run("init")
    run("config", "user.email", "test@example.com")
    run("config", "user.name", "Test")
    // write a file, stage, commit...
    return dir
}

Testing getAheadBehind requires a bare remote and a clone:

bare := t.TempDir()
exec.Command("git", "init", "--bare", bare).Run()
clone := t.TempDir()
exec.Command("git", "clone", bare, clone).Run()

Service layer: git.Service, OnStartup, handleQuery, and handleTask depend on framework.Core. Test these indirectly through DirtyRepos(), AheadRepos(), and Status() by setting lastStatus directly, or via integration tests.

agentci/ — unit tests only

agentci/ functions are pure (no I/O except SSH exec construction) and test without mocks:

func TestSanitizePath_Good(t *testing.T) {
    result, err := agentci.SanitizePath("myrepo")
    require.NoError(t, err)
    assert.Equal(t, "myrepo", result)
}

SanitizePath("../secret") returns "secret" — it strips the directory component via filepath.Base rather than rejecting the input. This is the documented, correct behaviour.

jobrunner/ — table-driven handler tests

Handler tests use the JobHandler interface directly with a mock forge.Client constructed via httptest. The preferred pattern is table-driven:

tests := []struct {
    name   string
    signal *jobrunner.PipelineSignal
    want   bool
}{
    {"merged PR", &jobrunner.PipelineSignal{PRState: "MERGED"}, true},
    {"open PR", &jobrunner.PipelineSignal{PRState: "OPEN"}, false},
}
for _, tt := range tests {
    t.Run(tt.name, func(t *testing.T) {
        assert.Equal(t, tt.want, handler.Match(tt.signal))
    })
}

Integration tests in handlers/integration_test.go test the full signal-to-result flow: construct a Poller, register sources and handlers, call RunOnce, verify journal entries and Forgejo API calls on the mock server.

collect/ — mixed unit and HTTP mock

Pure functions (state management, rate limiter logic, event dispatch) test without I/O. HTTP-dependent collectors (Collect methods for BitcoinTalk, GitHub, IACR, arXiv) require mock HTTP servers:

srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
    w.Header().Set("Content-Type", "application/json")
    json.NewEncoder(w).Encode(mockResponse)
}))
t.Cleanup(srv.Close)

Coding Standards

Language

Use UK English throughout: colour, organisation, centre, licence (noun), authorise, behaviour. Never American spellings.

Go style

  • All parameters and return types must have explicit type declarations.
  • Error strings follow "package.Function: context: %w" for wrapped errors, "package.Function: message" for sentinel errors. No bare fmt.Errorf("something failed") strings.
  • Import groups: stdlib, then forge.lthn.ai/..., then third-party, each separated by a blank line.
  • Use testify/require for fatal assertions, testify/assert for non-fatal. Prefer require.NoError over assert.NoError when subsequent test steps depend on the result.
  • Test naming convention: _Good (happy path), _Bad (expected error), _Ugly (panic/edge case).

Error wrapping

// Correct — contextual prefix with package.Function
return nil, fmt.Errorf("forge.CreateRepo: marshal options: %w", err)

// Correct — using the log.E helper from core/go
return nil, log.E("forge.CreateRepo", "failed to create repository", err)

// Incorrect — bare error with no context
return nil, fmt.Errorf("failed")

Context propagation

  • git/ and collect/ propagate context correctly via exec.CommandContext.
  • forge/ and gitea/ accept context at the wrapper boundary but cannot pass it to the SDK (SDK limitation, see architecture.md).
  • agentci/ uses SecureSSHCommandContext for all SSH operations — never use SecureSSHCommand (deprecated).

Commit Conventions

Use conventional commits with a package scope:

feat(forge): add GetCombinedStatus wrapper
fix(jobrunner): prevent double-dispatch on in-progress issues
test(git): add ahead/behind with bare remote
docs(agentci): document Clotho dual-run flow

Valid types: feat, fix, test, docs, refactor, chore.

Every commit must include the co-author trailer:

Co-Authored-By: Virgil <virgil@lethean.io>

Licence

EUPL-1.2. All source files must carry the EUPL-1.2 licence header if one is added to the project. The licence is compatible with GPL v2/v3 and AGPL v3.


Remote and Push

The canonical remote is on Forgejo via SSH:

git push origin main
# Remote: ssh://git@forge.lthn.ai:2223/core/go-scm.git

HTTPS authentication to forge.lthn.ai is not configured — always use SSH. The SSH port is 2223.


Adding a New Package

  1. Create the package directory under the module root.
  2. Add package <name> with a doc comment describing the package's purpose.
  3. Follow the existing client.go / config.go / types.go naming pattern where applicable.
  4. Write tests from the start — avoid creating packages without at least a skeleton test file.
  5. Add the package to the dependency graph in docs/architecture.md.
  6. Import groups must be maintained: stdlib, then forge.lthn.ai/..., then third-party.

Adding a New Handler

  1. Create jobrunner/handlers/<name>.go with a struct implementing jobrunner.JobHandler.
  2. Name() returns a lowercase identifier (e.g. "tick_parent").
  3. Match(signal) should be narrow — handlers are checked in registration order and the first match wins.
  4. Execute(ctx, signal) must always return an *ActionResult, even on partial failure.
  5. Add a corresponding <name>_test.go with at minimum one _Good and one _Bad test.
  6. Register the handler in Poller configuration alongside existing handlers.