flue dev exits when config watcher hits EMFILE/ENOSPC

[kalepail]

Summary

flue dev can fail before starting the local development server when the parent process cannot create its config-directory watcher:

Error: Config watcher failed: EMFILE: too many open files, watch

This is especially likely in large repos, repos with many generated files, or developer machines that already have watcher pressure from editors, agents, test runners, Vite/Wrangler, or other dev servers.

Why this matters

The failure happens in the parent flue dev supervisor while watching the resolved config file directory. In the common case where flue.config.* lives at the project root, this means a single failed fs.watch(root) aborts the entire dev command before the actual Node/Cloudflare local server can run.

That blocks the normal documented development path:

• flue dev does not reach the app server startup.
• Cloudflare-target apps never get to Vite/Wrangler.
• Workarounds like running generated Wrangler output directly bypass Flue's normal dev supervisor and are not equivalent to the intended development flow.

Observed reproduction

Downstream, on a Cloudflare-target Flue app:

./node_modules/.bin/flue dev --target cloudflare --port 3593

failed immediately with:

Error: Config watcher failed: EMFILE: too many open files, watch

Instrumentation of fs.watch showed the first failed watch target was the project root:

[debug-fs-watch] #1 /path/to/project
Error: Config watcher failed: EMFILE: too many open files, watch

After a fallback patch, the same normal command started successfully and served /health.

Expected behavior

If the config-directory watcher fails specifically because of watcher resource exhaustion (EMFILE or ENOSPC), flue dev should degrade gracefully instead of aborting.

Since the supervisor already has the exact resolved config file list, it can safely poll those specific files with fs.watchFile() as a fallback while preserving config restart behavior.

Non-resource watcher errors should still fail fast.

Proposed fix

PR #313 implements this:

• catches synchronous fs.watch(directory) failures for config directories;
• catches async watcher error events;
• falls back to polling the resolved config files for EMFILE / ENOSPC;
• keeps the current fail-fast behavior for unrelated watcher errors;
• adds a regression test that forces fs.watch(projectRoot) to throw EMFILE, verifies flue dev still starts, and verifies a config edit still restarts the dev session.

Related PR

Fix proposed in:

• Handle config watcher resource exhaustion in flue dev #313

[kalepail]

Closing this as a duplicate of #314. #314 is the auto-created triage issue from PR #313 and should be the canonical thread for the flue dev EMFILE/ENOSPC config watcher fallback.