Support async hook routes and bind the request for FastAPI hook routes

[T4rk1n]

Summary

Hook routes (dash.hooks.route) didn't work as async def, and on the FastAPI backend they couldn't read the request at all. The same FastAPI issue also broke MCP (_mcp).

Found while porting the Plotly Cloud dev tools publish route to the Quart/FastAPI backends (plotly/plotly-cloud-extension#112), which currently needs workarounds for both.

Changes

• Async hook routes: with_app_context_factory wrapped every hook route in a sync function, so an async def route returned 500 on Flask and Quart and failed on FastAPI. It now returns an async wrapper for coroutine functions (Flask needs flask[async]).
• App context leak: both wrappers set app_context in the caller's context, so dash.get_app() kept returning that app on the calling thread after the request. The sync wrapper now sets it only inside the copied context; the async wrapper sets it in the awaiting task's context and resets it afterwards. (Setting it in a copied context doesn't work for coroutines: ctx.run(func) only creates the coroutine there, and its body runs in the awaiting task's context.)
• FastAPI middleware: it only bound the current request for _dash-* paths, so request_adapter() raised "No active request in context" in hook routes and in MCP. It now also handles exact-match routes registered through Dash (app.routes).
• FastAPI body replay: the middleware reads the JSON body up front, which left the body stream empty for a handler reading it itself (the request hung). It now replays the body it read to the handler.
• Tests: hook_cleanup reset hooks._ns["route"] instead of "routes", so hook routes leaked between tests.

Test plan

• New tests/backend_tests/test_hook_routes.py: sync and async hook routes on Flask, Quart and FastAPI, reading JSON through the request adapter; get_app() returns the app serving the request with two apps; a FastAPI handler reading its own body. 4 of these fail without the fix.
• tests/backend_tests + tests/integration/test_hooks.py: 47 passed, 1 skipped
• tests/unit: 716 passed; the 3 failures also fail on dev locally (no Firefox, flask-compress installed, port 8050 subprocess test)
• black 22.3.0, flake8, pylint clean on changed files
• CI

🤖 Generated with Claude Code

[sonarqubecloud]

Quality Gate passed

Issues
0 New issues
0 Accepted issues

Measures
0 Security Hotspots
0.0% Coverage on New Code
0.0% Duplication on New Code

See analysis details on SonarQube Cloud

[github-actions]

Dash performance benchmarks

⚠️ regressions to review

	scenario	metric	p90 (ms)	median	growth	baseline p90	note
⚠️	full_children_replace	replace_ms	4365.6	1549.2	19.78x	4697.1	1.4x baseline (norm)
✅	callback_chain	chain_ms	325.7	314.9	0.92x	499.1
✅	callback_chain	graph_ms	1.6	1.6	1.0x	2.4
✅	callback_fanout	fanout_ms	64.5	60.6	0.92x	92.5
✅	deep_nesting	render_ms	41.0	39.1	0.95x	56.8
✅	initial_render_large	render_ms	495.9	417.9	1.12x	694.4
✅	initial_render_small	render_ms	71.6	63.2	1.0x	104.0
✅	patch_append_nested	append_ms	102.6	72.1	2.34x	192.3
✅	patch_append_toplevel	append_ms	93.9	64.1	2.16x	140.2
✅	patch_scalar_update_large	update_ms	117.9	107.6	1.0x	202.9
✅	wildcard_all_resolve	wildcard_ms	217.3	212.2	0.97x	313.6
✅	wildcard_all_resolve	graph_ms	0.8	0.8	1.0x	1.3

growth = late-third / early-third per-op time; ~1 is flat, a large value means the per-op cost scales with accumulated state.

machine scale vs baseline: 0.66x - divided out of the baseline ratios so they compare like for like (the absolute warn/fail ceilings are left un-scaled); calibrated on initial_render_small.