Use service name as service worker script name

[mrbbot]

Hey! 👋 Currently, when multiple nanoservices are configured with service worker scripts, they'll all use worker.js as their script origin. This means if an error is thrown, it's difficult to identify which script threw from inspecting the stack trace. We hit this issue in Miniflare, when attempting to locate source maps for service workers: https://github.com/cloudflare/miniflare/blob/5ecaae2b482feb583d8eeca2310785ea4d68165c/packages/tre/src/plugins/core/prettyerror.ts#L37-L41.

This PR adds a new name option for configuring the script origin of service worker scripts. Rather than making serviceWorkerScript a :group with name, I've added a new serviceWorker union member to maintain text-format compatibility.

[kentonv]

If providing a filename is really the right approach, this seems like the best way to do it.

But I have to wonder about this:

This means if an error is thrown, it's difficult to identify which script threw from inspecting the stack trace.

If there's only one script, why is it difficult to identify?

Is the problem really that separate services are not distinguished from each other? If so, would an alternative option be to bake the service name into the script's display name? That could happen automatically and would benefit everyone.

If that solves the problem for miniflare, and miniflare is then able to apply source mapping to map locations back to their original input source code, then do we still need to add this filename option? I just feel like no one would ever bother specifying it when writing a worker by hand, and tool-generated configs don't really need it if they are applying source mapping on top anyway.

[mrbbot]

That's a good point.

Is the problem really that separate services are not distinguished from each other?

Yep, this is the problem. Using <worker_name>(.js) as the origin would be enough for us to distinguish services.

However, for users that are just using workerd directly, having the ability to specify the file name might be nice, as clicking on stack trace call sites in IDEs often takes you directly to the code, assuming the names match. As you say though, it doesn't seem likely people would bother specifying this in embed and then again as (file)name. Would definitely prefer if we didn't need two ways of defining service worker scripts too.

[mrbbot]

Ok, I've switched to using the service name as the script name. This is sufficient for Miniflare to distinguish scripts. If users want IDE click-to-location in stack traces, they can set the service name to their filename. 👍 This needs a small internal change too, so I'll make the PR for that now.

[jasnell]

LGTM but let's also have @kentonv review.

[jasnell]

@mrbbot ... I left a comment on the internal repo also but the CI run for this is fairly old. Can I ask you to rebase this and the internal repo so we can get a fresh CI run before we merge this.

[mrbbot]

Sure! I'm away from my laptop for the next couple days, but will do this when I get back. 👍