perf(dev): faster cold-start for dev web serving

[t3dotgg]

Problem

Opening a dev build for the first time on a fresh worktree hangs on the splash screen for minutes while CPU sits idle. Cold-start is latency-bound, not compute-bound: unbundled dev serves every module as its own request, discovered one import level at a time, and over a tailnet origin each level pays a network round trip. A fresh worktree also starts with a cold dependency-optimizer cache, and the server does no transform work until the browser asks.

Solution

Four independent fixes, one per stall:

• --share defaults to bundled dev. The dev runner now sets T3CODE_BUNDLED_DEV=1 for shared runs, collapsing thousands of per-module requests into a few chunks. It only defaults the variable, so T3CODE_BUNDLED_DEV=0 still opts a shared run back out. Local (non-shared) dev is unchanged.
• server.warmup on the web entry. Vite transforms the whole module graph at server start instead of on first request, so first paint pays transfer, not discover-then-transform.
• Worktree setup pre-warms the dep cache. The t3.json setup script now runs apps/web/scripts/warm-dep-cache.ts after vp i, so the first dev start finds a valid node_modules/.vite/deps instead of running the optimize pass while the page waits. The cache can't be shared between worktrees (Vite's config hash includes the absolute root), so each worktree warms its own; re-running is a fast no-op.
• Dev-served JS is brotli-compressed. The dev server sent JS uncompressed, which localhost never notices but a remote browser pays for in full — bundled dev's ~25 MB chunk took ~55 s over a real uplink. A dev-only compression middleware (brotli quality 5) moves the same chunk as ~4.3 MB, cutting remote cold load to seconds for a few hundred ms of encode time.

Verified locally: warm script populates the cache from clean in ~1s on a fresh worktree and the dev server reuses it without re-optimizing; bundled dev + warmup serve the entry and chunk bundle without errors; measured the served chunk at 4.3 MB (br) vs 25 MB (identity) over the tailnet; dev-runner tests cover the new default and the opt-out.

---

Change made by Claude Fable 5 via Claude Code.

🤖 Generated with Claude Code

Note

Reduce cold-start latency for dev web serving with compression and pre-warming

• Adds a Brotli compression middleware plugin (vite.config.ts) to the Vite dev server via the compression package at quality level 5.
• Adds server.warmup with ./src/main.tsx as the entry so Vite pre-transforms the module graph on startup, reducing first-request latency.
• Adds a warm-dep-cache.ts script that manually invokes Vite's optimizeDeps to pre-warm the dependency optimizer cache before the server starts.
• Sets T3CODE_BUNDLED_DEV=1 by default for --share runs in dev-runner.ts, with opt-out via T3CODE_BUNDLED_DEV=0.

^{Macroscope summarized 4c0dabd.}

---

Note

Low Risk
Dev-only tooling (Vite config, worktree setup script, dev-runner env defaults); no production auth, data, or API behavior changes.

Overview
Cuts minutes-long splash hangs on fresh worktrees and remote vp run dev --share sessions by attacking network-bound cold start, not CPU.

vp run dev --share now defaults T3CODE_BUNDLED_DEV=1 so remote browsers get a few bundled chunks instead of a per-module tailnet waterfall; T3CODE_BUNDLED_DEV=0 still opts out. Local non-shared dev is unchanged.

The web Vite dev server adds Brotli compression (quality 5) on serve, server.warmup for ./src/main.tsx so transforms run at startup, and documents the bundled-dev default in config comments.

Worktree setup (t3.json) runs apps/web/scripts/warm-dep-cache.ts after install to pre-fill node_modules/.vite/deps per worktree (re-run is a no-op when warm).

Tests cover the share/bundled-dev env default and opt-out; internals docs note the --share behavior.

^{Reviewed by Cursor Bugbot for commit 4c0dabd. Bugbot is set up for automated code reviews on this repo. Configure here.}

[coderabbitai]

Important

Review skipped

Auto reviews are disabled on this repository. Please check the settings in the CodeRabbit UI or the .coderabbit.yaml file in this repository. To trigger a single review, invoke the @coderabbitai review command.

⚙️ Run configuration

Configuration used: Repository UI

Review profile: CHILL

Plan: Pro Plus

Run ID: 2f11a08b-3501-4fec-bf9f-a6fb834336d6

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

• 🔍 Trigger review

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}

[github-actions]

Thread transfer impact

✅ Thread transfer remains within every enforced ceiling.

Provider	Metric	Main baseline	This PR	Impact	PR ceiling
Codex	Total thread wire	11.3 KiB	11.3 KiB	−15 B (−0.1%)	15.1 KiB	✅
Codex	Thread snapshot wire	5.5 KiB	5.4 KiB	−2 B (−0.0%)	7.3 KiB	✅
Codex	Live turn WebSocket wire	5.9 KiB	5.9 KiB	−13 B (−0.2%)	7.8 KiB	✅
Codex	Live turn WebSocket decoded	49.7 KiB	49.7 KiB	0 B (0.0%)	66.4 KiB	✅
Codex	Live turn messages	16	16	0 (0.0%)	21	✅
Claude	Total thread wire	11.3 KiB	11.3 KiB	−5 B (−0.0%)	15.1 KiB	✅
Claude	Thread snapshot wire	5.5 KiB	5.5 KiB	−5 B (−0.1%)	7.3 KiB	✅
Claude	Live turn WebSocket wire	5.9 KiB	5.9 KiB	0 B (0.0%)	7.8 KiB	✅
Claude	Live turn WebSocket decoded	50.6 KiB	50.6 KiB	0 B (0.0%)	66.4 KiB	✅
Claude	Live turn messages	16	16	0 (0.0%)	21	✅

Baseline: 23f0a1a · PR result: 4c0dabd · Source CI: success

Scenario and decoded snapshot size

10 historical turns, 5 command tools per turn, 878.9 KiB retained MCP result per historical turn, and a 1.05 MiB retained result in the measured turn.

• Codex decoded thread snapshot: 94.6 KiB
• Claude decoded thread snapshot: 95.3 KiB

Updated in place by a trusted workflow. PR artifacts are strictly validated and never executed.

[macroscopeapp]

Approvability

Verdict: Approved 4c0dabd

Development-only performance optimization for faster cold-starts. All changes are scoped to dev server (compression plugin uses apply: "serve"), worktree setup, and dev runner behavior. No production code paths affected, comprehensive tests added, and well-documented.

^{You can customize Macroscope's approvability policy. Learn more.}

Dismissing prior approval to re-evaluate 42e84d8

[t3dotgg]

Added a third improvement after live testing over the tailnet: the dev server now brotli-compresses served JS (quality 5). Bundled dev's ~25 MB chunk was taking ~55 s on a real uplink; it now transfers as ~4.3 MB, cutting remote cold load to seconds.

🤖 Generated with Claude Code