opencode-config/tests/_ts_loader.mjs
Sergey 1db2c800d4
feat(tools): add commit, create-pr, create-issue tools (#38)
* feat(tools): add commit tool with validation

* feat(tools): add create-pr tool with validation

* feat(tools): add create-issue tool with validation

* test(tools): add tests for validation tools

* docs(handoff): add handoff and ADR for validation tools

* docs(handoff): set PR number

* docs: update project map + fix ADR-015 typo

---------

Co-authored-by: opencode-agent <agent@slaid098.dev>
2026-07-24 04:17:44 +03:00

238 lines
No EOL
10 KiB
JavaScript

// Test harness for .opencode/tools/pipeline-status.ts.
//
// Why this file: there is no bun/tsx/esbuild on the CI runner (only node +
// pytest). The pipeline-status.ts tool is a TypeScript module (Bun runtime
// for opencode, not runnable with plain `node`). To exercise the tool's
// execute() function in pytest, we strip TS-only syntax (import type
// annotations + ESM import -> CJS require) and load the resulting JS as a
// CommonJS module. The tool() factory from @opencode-ai/plugin is identity
// (returns its argument), so we can replace it with a passthrough shim
// without changing the tool semantics.
//
// This file is consumed by tests/test_pipeline_status_tool.py via:
// node tests/_ts_loader.mjs <mode> [args...]
// where <mode> is one of: load, exec, exec_real.
//
// - load: print the {description, args keys} of the tool — sanity check.
// - exec: call execute({pr_number: <int>}) with a stubbed spawnSync that
// returns whatever argv it received, plus a fixture stdout. Used by unit
// tests to verify args passed and output trimming.
// - exec_real: call execute({pr_number: <int>}) with the REAL spawnSync,
// used by the integration test against the real pipeline-status.py.
import { readFileSync } from "node:fs"
import { fileURLToPath } from "node:url"
import path from "node:path"
import { spawnSync } from "node:child_process"
const REPO_ROOT = path.resolve(fileURLToPath(import.meta.url), "..", "..")
const DEFAULT_TS_FILE = path.join(REPO_ROOT, ".opencode", "tools", "pipeline-status.ts")
// Parameterize via env var TS_FILE (relative to REPO_ROOT) so other TS tool
// wrappers (e.g. spec-status.ts) can be loaded by the same harness without
// breaking existing callers that don't set TS_FILE (default: pipeline-status.ts).
const TS_FILE = process.env.TS_FILE
? path.resolve(REPO_ROOT, process.env.TS_FILE)
: DEFAULT_TS_FILE
// Minimal zod shim covering the methods the tool actually uses:
// tool.schema.number().describe("...")
// At runtime (in opencode/Bun) `tool.schema` is the real zod. Here we
// only need a chainable builder that returns an object with .describe()
// — no validation occurs in execute().
function makeZodShim() {
const chain = () => {
const obj = {
describe() { return obj },
optional() { return obj },
// Add more methods as future tool definitions need them.
}
return obj
}
return {
number: chain,
string: chain,
boolean: chain,
array: chain,
object: chain,
}
}
const zodShim = makeZodShim()
function stripTs(src) {
// Minimal TS -> JS for this specific file:
// 1) `import { spawnSync } from "child_process"` -> `const { spawnSync } = require("child_process")`
// 2) `import { tool } from "@opencode-ai/plugin"` -> `const tool = (x) => x`
// 3) `import path from "path"` -> `const path = require("path")`
// 4) `import.meta.dir` -> a stub pointing at .opencode/tools (so the script
// path resolves to .opencode/scripts/pipeline-status.py)
// 5) `args: z.ZodObject` -> the args are referenced inside execute as
// `args.pr_number`; the schema itself is unused at runtime here.
// 6) Strip `: type` annotations and `async execute(args)` stays.
let out = src
out = out.replace(/^import\s+\{\s*spawnSync\s*\}\s+from\s+["']child_process["'];?\s*$/m, 'const { spawnSync } = require("child_process");')
out = out.replace(/^import\s+path\s+from\s+["']path["'];?\s*$/m, 'const path = require("path");')
out = out.replace(/^import\s+\{\s*tool\s*\}\s+from\s+["']@opencode-ai\/plugin["'];?\s*$/m, 'const tool = (x) => x;')
// Replace `import.meta.dir` with the directory of the TS file.
out = out.replace(/import\.meta\.dir/g, JSON.stringify(path.dirname(TS_FILE)))
return out
}
function loadTool(spawnSyncImpl) {
// Provide a CommonJS module sandbox so the tool file's `export default`
// becomes accessible via `module.exports.default`.
let src = stripTs(readFileSync(TS_FILE, "utf-8"))
// Strip the require/const declarations we replace with sandbox args, so
// we don't get "Identifier already declared" between Function args and
// the in-source `const { spawnSync } = require(...)` lines.
src = src.replace(/^const\s+\{\s*spawnSync\s*\}\s*=\s*require\(["']child_process["']\);?\s*$/m, "")
src = src.replace(/^const\s+path\s*=\s*require\(["']path["']\);?\s*$/m, "")
src = src.replace(/^const\s+tool\s*=\s*\(x\)\s*=>\s*x;?\s*$/m, "")
// Convert `export default tool({...})` into `module.exports.default = tool({...})`
const cjs = src.replace(/^export default /m, "module.exports.default = ")
// tool shim with .schema = zodShim (since pipeline-status.ts uses tool.schema.number())
const toolShim = (x) => x
toolShim.schema = zodShim
const fn = new Function(
"module",
"require",
"spawnSync",
"path",
"tool",
cjs + "\nreturn module.exports.default;",
)
return fn({ exports: {} }, (name) => {
if (name === "child_process") return { spawnSync: spawnSyncImpl }
if (name === "path") return path
throw new Error("unexpected require: " + name)
}, spawnSyncImpl, path, toolShim)
}
function buildExecArgs(tool, rawValue) {
// Interpret ``rawValue`` (argv string) as the tool's first declared arg.
// - pipeline-status.ts: ``pr_number`` (int) → parseInt
// - spec-status.ts: ``validate`` (bool) → /true/i match
// Detection is dynamic so the harness works for any single-arg tool
// without hardcoding tool names. If the tool declares no args, return {}.
const keys = Object.keys(tool.args || {})
if (keys.length === 0) return {}
const first = keys[0]
if (first === "pr_number") return { pr_number: parseInt(rawValue, 10) }
if (first === "validate") return { validate: /^true$/i.test(rawValue || "") }
// Fallback heuristic: numeric → int, else bool-ish.
return { [first]: rawValue }
}
// Multi-arg tools (commit.ts, create-pr.ts, create-issue.ts) declare several
// args (message, title, body, issue_number, labels). The single-arg
// ``buildExecArgs`` can't handle them. ``exec_stub_json`` mode passes the
// full args object as a JSON string + a JSON array of sequential stub
// responses (one per spawnSync call — commit.ts makes 2: git diff, git commit).
function buildExecArgsFromJson(rawValue) {
return JSON.parse(rawValue)
}
function buildStubSequencer(responses) {
// Return a stub function that returns responses[callIndex] for each call,
// cycling through the list if there are more calls than responses.
let idx = 0
return (cmd, args, opts) => {
const r = responses[idx % responses.length]
idx++
return { status: r.status, stdout: r.stdout ?? "", stderr: r.stderr ?? "" }
}
}
function main() {
const mode = process.argv[2]
if (!mode) {
console.error("usage: node _ts_loader.mjs <mode> [args...]")
process.exit(2)
}
if (mode === "load") {
const t = loadTool(spawnSync)
console.log(JSON.stringify({ description: t.description, args: Object.keys(t.args) }))
return
}
if (mode === "exec_stub") {
// Args: <first_arg_value> <stub_status> <stub_stdout> <stub_stderr>
// The first arg is interpreted based on which tool is loaded:
// - pipeline-status.ts declares ``pr_number`` (int)
// - spec-status.ts declares ``validate`` (bool)
// Detected dynamically from ``t.args`` keys so the harness stays generic.
const stubStatus = parseInt(process.argv[4], 10)
const stubStdout = process.argv[5]
const stubStderr = process.argv[6] || ""
const callLog = []
const stub = (cmd, args, opts) => {
callLog.push({ cmd, args, opts })
return { status: stubStatus, stdout: stubStdout, stderr: stubStderr }
}
const t = loadTool(stub)
const execArgs = buildExecArgs(t, process.argv[3])
t.execute(execArgs, {
// ToolContext — only fields the tool actually touches. Our tool
// touches none of the ctx fields, so this can be empty-ish.
sessionID: "test", messageID: "test", agent: "test",
directory: REPO_ROOT, worktree: REPO_ROOT,
abort: new AbortController().signal,
metadata() {}, async ask() {},
}).then(
(result) => {
console.log(JSON.stringify({ result, calls: callLog }))
},
(err) => {
console.log(JSON.stringify({ error: String(err), calls: callLog }))
},
)
return
}
if (mode === "exec_stub_json") {
// Multi-arg tools (commit.ts, create-pr.ts, create-issue.ts).
// Args: <args_json> <responses_json>
// args_json — JSON string of the args object, e.g. {"message":"feat(x): y"}
// responses_json — JSON array of {status, stdout, stderr} stub
// responses, returned sequentially per spawnSync call. Tools that
// make N spawnSync calls need N entries (extra calls cycle back).
const execArgs = buildExecArgsFromJson(process.argv[3])
const responses = JSON.parse(process.argv[4])
const callLog = []
const stub = (cmd, args, opts) => {
callLog.push({ cmd, args, opts })
const r = responses[callLog.length - 1] || responses[responses.length - 1]
return { status: r.status, stdout: r.stdout ?? "", stderr: r.stderr ?? "" }
}
const t = loadTool(stub)
t.execute(execArgs, {
sessionID: "test", messageID: "test", agent: "test",
directory: REPO_ROOT, worktree: REPO_ROOT,
abort: new AbortController().signal,
metadata() {}, async ask() {},
}).then(
(result) => {
console.log(JSON.stringify({ result, calls: callLog }))
},
(err) => {
console.log(JSON.stringify({ error: String(err), calls: callLog }))
},
)
return
}
if (mode === "exec_real") {
const t = loadTool(spawnSync) // use real spawnSync
const execArgs = buildExecArgs(t, process.argv[3])
t.execute(execArgs, {
sessionID: "test", messageID: "test", agent: "test",
directory: REPO_ROOT, worktree: REPO_ROOT,
abort: new AbortController().signal,
metadata() {}, async ask() {},
}).then(
(result) => console.log(JSON.stringify({ result })),
(err) => console.log(JSON.stringify({ error: String(err) })),
)
return
}
console.error("unknown mode: " + mode)
process.exit(2)
}
main()