feat: Add basePath support on web app - Closes #7624

[thebreiflabb]

Allow serving web app within a subPath of a host

• Make app build with relative basepath
• Initialize app router with basepath if defined from server
• Add --base-path CLI param and server.basePath option
• Remove basePath in Server before processing routing
• Add basePath in requestContext for easy access in handlers
• Normalize CSP on nonce for easily handling inline scripts
• Use built-in Bun HTMLRewriter for manipulating HTML basepath before serving to client

Issue for this PR

Closes #7624

Type of change

• Bug fix
• New feature
• Refactor / code improvement
• Documentation

What does this PR do?

Add support for serving the web app under a basePath. This is useful when serving OpenCode on web through an ingress / proxy setup, where you might have https://www.mydomain.com/apps/opencode/

How to test:

(dev still works as before, but the new basePath functionality only works in built binary until the new app is deployed to https://app.opencode.ai)

bun packages/opencode/script/build.ts --single

./packages/opencode/dist/opencode-linux-x64/bin/opencode web --base-path /my-base/sub-path/

Will open on http://127.0.0.1:4096/my-base/sub-path/

Alternatively, configure opencode.json with:

{
  "server": {
    "basePath": "/my-base/sub-path/"
  }
}

---

This is solved with combining changes in:

Web App

1. Configure Vite to build with relative basePath, where assets will load relatively from the current path
2. Load __OPENCODE__.basePath if defined from the server, use to initialize Solid Router and as base for API calls to server

Opencode Server app

1. Add server.basePath config option and --base-path CLI argument
2. Configure Server listener to remove basePath before Hono app route handling, making it transparent /my-base/path/global/event => /global/event
3. To avoid complicating the Hono setup, passing around Hono<Env> types, a simple AsyncLocalStorage - requestContext holds the basePath context to use for nested handlers
4. When serving the HTML, Bun's built-in HTMLRewriter handles:
   • Rewriting of the basePaths on links in the HTML
   • Adding script nonce
   • Add basePath to/initialize window.__OPENCODE__
5. Switch to using a nonce for CSP, to simplify multiple inline scripts being allowed in the CSP

How did you verify your code works?

I have tested built binary:

./script/build.ts --single

With no basePath, --base-path /various/levels/ as well as through config server.basePath option. This looks to work very well

---

When running bun dev, this works quite well, and does not break, however since this proxies to https://app.opencode.ai - the referenced JS assets are not updated with the new basePath handling, and the basePath is ignored until this app is deployed with these changes (works as if there is no basePath defined)

Screenshots / recordings

If this is a UI change, please include a screenshot or recording.

Checklist

• I have tested my changes locally
• I have not included unrelated changes in this PR

[github-actions]

The following comment was made by an LLM, it may be inaccurate:

Potential duplicate PRs found:

1. PR feat: base path support #7625 - "feat: base path support"

   • feat: base path support #7625
   • This appears to be directly related to the basePath feature and may be a prior or related attempt at implementing the same functionality.

2. PR feat: App - Support setting base URL during build #18209 - "feat: App - Support setting base URL during build"

   • feat: App - Support setting base URL during build #18209
   • This relates to base URL configuration for the app, which overlaps with the basePath initialization mentioned in the current PR.

Why they might be related:

• PR feat: base path support #7625 explicitly mentions "base path support" which is the core feature of PR feat: Add basePath support on web app - Closes #7624 #21243
• PR feat: App - Support setting base URL during build #18209 addresses base URL configuration for the app build, which is one component of the basePath feature being added

I recommend checking the status and content of these PRs to see if they've been superseded, are still open, or contain overlapping work.

[thebreiflabb]

Not sure why the test failed after rebasing, I don't believe the code has a touchpoint on the failing test, and tests are passing locally. Could it be a flaky test?