Move js-client bootstrapping to FastAPI

[guerler]

Requires #17507. Resolves #21153.

This replaces the entire historical bootstrap path with a single, explicit initialization flow.

The old layers disappear in full: Mako injected config, ad-hoc script tags, incremental config.set merges, RxJS config debouncing, init queues, defaultAppFactory, standardInit, and the monitor plumbing. None of these participate anymore.

Startup is reduced to one request and one construction step.

App Construction

The client issues a single GET to /context. That response contains the startup payload: config, detailed user, and a session CSRF token. No templates insert state. No merging or patching occurs afterward. initGalaxyInstance takes the JSON, builds a GalaxyApp, registers it, and installs the window.Galaxy getter. Construction is uniform across all entry points.

GalaxyApp options default deterministically. User and locale initialization no longer depend on template timing. No late mutation of config or user attributes remains. All states derive from the context payload and stay stable.

Elimination of Legacy Mechanism

The template based bootstrapping path is removed. JSAppLauncher is gone. Templates no longer try to prepare configuration or user objects. The JS bundle loads without server side state injection.

All RxJS sequencing is removed. There is no config observable, no debounce cycle, and no init queue. Nothing accumulates late additions. Initialization is synchronous except for the initial fetch.

New Mount Sequence

The mount sequence becomes linear and inspectable:

1. window load listener triggers.
2. initGalaxyInstance fetches /context and constructs GalaxyApp.
3. Router is created with access to GalaxyApp object.
4. Add-ons like Sentry and Webhooks are initialized with config and user available.
5. Vue app mounts.

No branching. No timing sensitivity. No template inserted mutations.

See: https://github.com/guerler/galaxy/blob/00fdf2556a473eb22ab16f91221c11c797b4ab2b/client/src/entry/analysis/index.ts

Configuration

All previous template delivered or JS injected configuration now arrives through /context: dynamic tool confs, activation and inactivity flags, webhooks enablement, analytics DSNs, workflow menu entries, and admin extended config. Consumers read from Galaxy.config and Galaxy.user or the Pinia stores, instead of inspecting template variables or legacy global state.

Result

Client startup collapses to a predictable, single source of truth. Every subsystem sees the same configuration and the same user payload. All historical timing and merging behavior is removed. The only contract the reviewer needs to understand is the /context response and the fixed initialization sequence.

Outlook

The context endpoint also fixes the mismatch between the initial bootstrap response and what the config and user Pinia stores expect. It already returns the exact fields produced by the regular config and user API endpoints, so the client gets the final, ready to use values on first load. That will let us fill both stores immediately without extra mapping or translation. If we later decide to fetch config and user separately, or to load parts of them on demand, nothing needs to change on the client because the context response is already built from those same endpoints. This keeps bootstrapping unified now while leaving room to break it apart cleanly later.

How to test the changes?

(Select all options that apply)

• I've included appropriate automated tests.
• This is a refactoring of components with existing test coverage.
• Instructions for manual testing are as follows:
  1. [add testing steps and prerequisites here if you didn't write automated tests covering all your changes]

License

• I agree to license these and all my past contributions to the core galaxy codebase under the MIT license.

[mvdbeek]

Design-wise I think it would be a good idea to make this as static as possible so we can heavily cache the response. fetch all the user-specific stuff later ?

[guerler]

Agreed, this is an initial very rough draft. We should remove unnecessary parts and keep it as minimal as possible. It may eventually fit into the configuration API, but it’s too early to decide.

[guerler]

@mvdbeek The PR is in decent shape now. Splitting the user and config pieces out, or adding caching, still makes sense, but we can handle that cleanly in a follow-up once these structural changes are in dev.

[jmchilton]

We've made so much of the API dynamic and user dependent and the config reloadable - so I'm not sure I see an easy path forward on this issue but I have created an issue for it here - #21373. Obviously this endpoint is on every critical path for the UI so even if the path isn't easy I'm sure it is possible to get some important speed up here if we are careful.

[davelopez]

There is a DetailedUserModel in lib/galaxy/schema/schema.py maybe you can use that to improve the API typing here.
Of course, you will need to add the new fields for the detailed view introduced in this PR in the model too :)

[guerler]

Switched to the DetailedUserModel.

[guerler]

@davelopez Switching to DetailedUserModel as in 52e8671 produces double encoding of the user id because of the serialization at: https://github.com/galaxyproject/galaxy/blob/dev/lib/galaxy/managers/users.py#L688. The user manager is used in several places, so removing encoding there is not guaranteed to be safe and would need broader review. This might fit better in the follow-up where config and user data move into their Pinia stores and the Galaxy.config and Galaxy.user attributes are likely to be removed.

[guerler]

Remaining test failures are unrelated. Will follow up with injecting and streamlining the user and config store contents in a separate PR.

[jmchilton]

The writeup is great and better than I was hoping even - thank you for writing that or prompting something to @guerler!

[jmchilton]

Other folks had more material critiques than me - so I'd love to get a second or third pair of eyes on this critical code but it seems like a very solid step in a a very solid direction to me.

[guerler]

Thank you for the review @jmchilton, @davelopez and @mvdbeek. Merging this now to move forward, and additional eyes are always welcome. I am happy to immediately address any follow up concerns.

[jmchilton]

Really nice work - thanks so much @guerler !