[yangshun]

When several routes match a URL, the most specific one now handles it, the same way in pages/, routes/ and WebSocket routes. Before, each scanner ranked overlapping routes differently, and some URLs went to the wrong file without a warning:

URL	Files	Was	Now
/users/about	pages/[lang]/about.tsx, pages/users/[id].tsx	[lang]/about	users/[id]
/users/new/edit	pages/users/[id]/edit.tsx, pages/users/new/[tab].tsx	users/[id]/edit	users/new/[tab]
/fr/about	pages/[lang]/about.tsx, pages/[lang]/[slug].tsx	[lang]/[slug]	[lang]/about
/fr/about	routes/[lang]/about.ts, routes/[category]/[item].ts	[category]/[item]	[lang]/about
/chat/lobby	routes/chat/[room].ws.ts, routes/chat/lobby.ws.ts	chat/[room]	chat/lobby

Pages compared whole pattern strings, where : sorts before letters, so any [param] beat a static segment at the same depth. This hit translated sites hardest: under pages/[lang]/ every page is dynamic, so [lang]/[slug] shadowed every static page beside it. WebSocket routes used plain string order. routes/ already compared segment by segment, but let parameter names break ties.

All three now share one comparator: catch-alls last, fully static routes first, then the first differing segment decides (static, then [param], then catch-all). Parameter names never decide. The Pages client matcher classifies segments with the same helper, so its speculative match agrees with the server.

Unchanged: catch-all routes still rank after every other route, so /docs/intro goes to [lang]/[slug] before docs/[...path].

Docs: a short "Overlapping routes" section in the server routing guide, which the pages guide points to.

Tests: a comparator test that checks each pair really overlaps (using the client regex) and that the order doesn't depend on input order, plus one regression test each for the pages scan (through the generated matchRoute), the routes/ scan and the WebSocket scan. All three scan tests fail on main.

🤖 Generated with Claude Code

[cpojer]

Love it.