[sdk/local-evaluation] getEnvironmentFlags() evaluations are not tracked for analytics

该 issue 指向 sdk/index.ts 中第 640 行附近的 getEnvironmentFlagsFromDocument，其中调用 Flags.fromEvaluationResult 时没有传入 analytics processor。请将其与第 669 行附近的 getIdentityFlagsFromDocument 比较，后者已经传入了 this.analyticsProcessor。运行 issue 中的复现脚本：当本地运行结果与远程运行一样报告 {"my_feature":5} 时即为完成。

Problem

With enableLocalEvaluation: true, flag evaluations from getEnvironmentFlags() are never counted towards analytics. The same code against the same environment reports usage correctly when local evaluation is off, so flag analytics are effectively unavailable to anyone evaluating locally.

The cause looks like a missed argument in getEnvironmentFlagsFromDocument:

// sdk/index.ts:640
const flags = Flags.fromEvaluationResult(evaluationResult);

The other three evaluation paths all pass the processor through:

• getIdentityFlagsFromDocument (local) -> sdk/index.ts:669, passes this.analyticsProcessor
• getEnvironmentFlagsFromApi (remote) -> sdk/index.ts:689
• getIdentityFlagsFromApi (remote) -> sdk/index.ts:710

Since the other local path already tracks, this doesn't look like a deliberate decision to skip analytics during local evaluation.

For what it's worth, the Python client passes it in the equivalent method, so the two SDKs behave differently here:

# flagsmith/flagsmith.py:535, _get_environment_flags_from_document
return Flags.from_evaluation_result(
    evaluation_result=evaluation_result,
    analytics_processor=self._analytics_processor,
    default_flag_handler=self.default_flag_handler,
)

Impact

getEnvironmentFlags() is the common path for server-side usage, and local evaluation is the recommended mode there, so this combination silently reports no usage at all. Nothing warns or errors, and the dashboard just shows the flag as unused. We ran into it while using analytics to decide which stale flags were safe to delete, where a flag wrongly showing zero evaluations is the dangerous direction to be wrong in.

Reproduction

Standalone script, no account needed; it stubs the API locally and compares the two modes.

const http = require('node:http');
const { Flagsmith } = require('flagsmith-nodejs');

const ENV_DOC = {
  api_key: 'ser.repro',
  project: { id: 1, name: 'repro', organisation: { id: 1, name: 'org', feature_analytics: true,
    stop_serving_flags: false, persist_trait_data: true }, hide_disabled_flags: false, segments: [] },
  feature_states: [{ feature: { id: 1, name: 'my_feature', type: 'STANDARD' },
    enabled: true, feature_state_value: null, featurestate_uuid: 'fs-1', multivariate_feature_state_values: [] }],
  identity_overrides: []
};
const API_FLAGS = [{ feature: { id: 1, name: 'my_feature', type: 'STANDARD' },
  enabled: true, feature_state_value: null }];

(async () => {
  const server = http.createServer((req, res) => {
    res.writeHead(200, { 'Content-Type': 'application/json' });
    if (req.url.includes('environment-document')) return res.end(JSON.stringify(ENV_DOC));
    if (req.url.includes('/flags')) return res.end(JSON.stringify(API_FLAGS));
    res.end('{}');
  });
  await new Promise(r => server.listen(0, '127.0.0.1', r));
  const apiUrl = `http://127.0.0.1:${server.address().port}/api/v1/`;

  async function run(local) {
    const f = new Flagsmith({ environmentKey: 'ser.repro', apiUrl,
      enableLocalEvaluation: local, enableAnalytics: true });
    const flags = await f.getEnvironmentFlags();
    for (let i = 0; i < 5; i++) await flags.isFeatureEnabled('my_feature');
    const buf = { ...(f.analyticsProcessor?.analyticsData ?? {}) };
    f.environmentDataPollingManager?.stop();
    console.log(`${local ? 'LOCAL ' : 'REMOTE'} -> buffer ${JSON.stringify(buf)}`);
  }

  await run(true);
  await run(false);
  server.close();
})();

Observed:

LOCAL  -> buffer {}
REMOTE -> buffer {"my_feature":5}

Expected: both report {"my_feature":5}.

Version: [email protected] (current npm latest), Node 24.13.0. Line numbers above are from main as of today, which still has the same call.

Workaround

Calling trackFeature(name) on the processor after each evaluation restores the counts, which is what we're doing for now.