feat: Add bun test runner integration

[Abhijeet Prasad (AbhiPrasad)]

resolves #1479

AI Summary

Adds initBunTestSuite — a Braintrust experiment integration for Bun's built-in test runner. Each suite.test() call registers a real bun:test test that logs inputs, outputs, and scorer results as experiment spans.

Usage

import { test, describe, afterAll } from "bun:test";
import { initBunTestSuite } from "braintrust";

describe("Translation Evaluation", () => {
  const suite = initBunTestSuite({
    projectName: "my-project",
    afterAll, // auto-flush when suite finishes
    test,     // bun's test function
  });

  suite.test("translates hello", {
    input: { text: "hello", lang: "es" },
    expected: "hola",
    scorers: [({ output, expected }) => ({
      name: "exact_match",
      score: output === expected ? 1 : 0,
    })],
  }, async ({ input }) => {
    return await translate(input.text, input.lang);
  });

  // untracked tests work alongside suite tests
  test("sanity check", () => expect(1 + 1).toBe(2));
});

Data expansion uses loops, and all bun:test modifiers are supported:

for (const [i, record] of cases.entries()) {
  suite.test(`case [${i}]`, { ...record, scorers }, async ({ input }) => translate(input));
}

suite.test.skip("wip", config, fn);
suite.test.failing("expected error", config, fn);
suite.test.skipIf(!process.env.CI)("CI only", config, fn);
// also: .only, .todo, .concurrent, .serial, .if, .todoIf

How it differs from wrapVitest and initNodeTestSuite

All three integrations share the same core (EvalConfig, ScorerFunction, runTracedEval) but adapt to each runner's API:

	wrapVitest	initNodeTestSuite	initBunTestSuite
Registration	bt.test(name, config, fn)	test(name, suite.eval(config, fn))	suite.test(name, config, fn)
Modifiers	skip, only, concurrent, todo	Inherited from node:test	skip, only, todo, failing, concurrent, serial, if, skipIf, todoIf
Extra APIs	logOutputs, logFeedback, expect wrapper	—	—

initBunTestSuite uses the suite.test(name, config, fn) shape because bun callbacks don't receive a context with .name — the test name must be captured at registration. This also makes modifier wrapping natural (suite.test.skip(...), suite.test.if(cond)(...)).

The test config parameter is generic (BunTestSuiteConfig<TTest>) so bun's actual Test<[]> type is forwarded without the SDK depending on bun's types. A runtime validateTestFunction() catches misconfiguration early. Modifiers like .only are bound lazily because bun's CI mode throws when test.only is even property-read.

Future extensions

• test.each() support for more ergonomic parameterized tests
• Custom bun reporter that captures results automatically without suite.test() wrappers
• logOutputs / logFeedback helpers for parity with the vitest integration
• Shared createTracedEvalFn to reduce duplication across runner integrations