app.py: add long-lived Cache-Control for large web/ assets via Starlette middleware

app.py

@@ -24,6 +24,8 @@ from collections import deque
 
 import gradio as gr
 
+from starlette.middleware.base import BaseHTTPMiddleware
+
 from lilyscript.generator import StreamingLilyletGenerator
 from lilyscript.postprocess import postprocess
 from lilyscript.mask_monitor import MaskMonitor, load_blacklist
@@ -45,6 +47,35 @@ EXAMPLES_DIR = os.path.join(HERE, 'examples')
 OUTPUT_DIR = os.path.join(HERE, 'outputs')
 WEB_DIR = os.path.join(HERE, 'web')		# vendored browser libs + score player (gitignored bundles)
 
+
+# Gradio serves our vendored web/ assets (soundfont gm.sf3 ~38 MB, verovio.bundle.js ~8 MB,
+# libfluidsynth ~2 MB, …) via `/gradio_api/file=<abs-path>` with NO Cache-Control header, so the
+# browser re-validates (or refetches) them on every load. These are large, static, and effectively
+# immutable per deploy — bundles carry a `?v=<mtime>` cache-buster (see _file_url), and the soundfont
+# rarely changes. This middleware adds a long-lived Cache-Control to responses for web/ assets so
+# returning visitors reuse them from the browser cache instead of re-downloading. It runs inside our
+# own app.py, so it applies on Hugging Face and ModelScope Spaces (where we are the server), unlike a
+# static-host CDN we couldn't configure.
+_WEB_ASSET_MARKER = '/web/'
+_CACHEABLE_EXTS = ('.sf3', '.sf2', '.js', '.wasm', '.css', '.mp3', '.ogg')
+
+
+class AssetCacheHeaderMiddleware(BaseHTTPMiddleware):
+	async def dispatch (self, request, call_next):
+		response = await call_next(request)
+		# Only touch Gradio's static-file route for our vendored web/ assets, and only if a
+		# Cache-Control wasn't already set. request.url.path holds the decoded `/gradio_api/file=...`.
+		path = request.url.path
+		if (
+			'/file=' in path
+			and _WEB_ASSET_MARKER in path
+			and path.endswith(_CACHEABLE_EXTS)
+			and 'cache-control' not in {k.lower() for k in response.headers.keys()}
+		):
+			# 30 days. Safe: bundle URLs change via ?v=<mtime> on redeploy; soundfont is stable.
+			response.headers['Cache-Control'] = 'public, max-age=2592000'
+		return response
+
 EXAMPLE_PREFIX = '\U0001F4C4 '		# 📄 examples
 OUTPUT_PREFIX = '✨ '			# ✨ session outputs
 
@@ -989,9 +1020,13 @@ def build_ui ():
 
 if __name__ == '__main__':
 	demo = build_ui()
+	from starlette.middleware import Middleware
 	demo.queue().launch(
 		theme=gr.themes.Soft(),
 		css=CUSTOM_CSS,
 		head=build_head(),
 		allowed_paths=[WEB_DIR],
+		# Inject a long-lived Cache-Control on our large static web/ assets (see
+		# AssetCacheHeaderMiddleware). app_kwargs flows into the FastAPI/Starlette constructor.
+		app_kwargs={'middleware': [Middleware(AssetCacheHeaderMiddleware)]},
 	)