Ensure api-resources are always loaded

[myieye]

Due to refactoring, the entry row loading skeletons now tend to get destroyed faster. And strangely enough even though they're skeletons, they trigger loading parts-of-speech from the back end. They're the first component to load parts of speech, which means the loading effect (inside the runed resource) is scoped to their lifecycle (this is a bad pattern that I hadn't noticed before. I wasted lots of time looking for a clean way to redesign stuff, but failed to find anything really elegant). Because they started getting destroyed sooner, they were getting destroyed before the parts of speech were even loaded, which meant that the parts of speech never got loaded for any component that needed them 🙀.

[coderabbitai]

📝 Walkthrough

Walkthrough

This PR refactors the resource initialization logic in the project context by replacing a throttling-based mechanism with a queuing system that defers resource initialization until the API is available. The change introduces a queue for managing pending resource callbacks and generates unique identifiers for tracking resources.

Changes

Cohort / File(s)	Summary
Project Context Refactoring frontend/viewer/src/project/project-context.svelte.ts	Replaced throttling-based initial load logic with a queuing mechanism that defers resource initialization until the API is set up; removed watchOnce import, added randomId import, introduced #queuedResources state queue, and implemented callback-based initialization on API availability.

Estimated code review effort

🎯 2 (Simple) | ⏱️ ~10 minutes

Suggested reviewers

• rmunn

Poem

🐰 wiggles whiskers thoughtfully
Queue the resources, don't rush ahead,
Wait for the API to wake from bed,
Then spring them forth with hoppy pride! 🌟

🚥 Pre-merge checks | ✅ 3

✅ Passed checks (3 passed)

Check name	Status	Explanation
Title check	✅ Passed	The title 'Ensure api-resources are always loaded' directly aligns with the main objective of fixing a resource loading issue where API resources were being destroyed before completion.
Description check	✅ Passed	The description explains the root cause (skeleton components destroyed sooner due to refactoring) and the impact (parts-of-speech loading was scoped to skeleton lifecycle), which is directly related to the changeset's solution of deferring resource initialization until the API is available.
Docstring Coverage	✅ Passed	No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check.

_{✏️ Tip: You can configure your own custom pre-merge checks in the settings.}

✨ Finishing touches

• 📝 Generate docstrings

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Post copyable unit tests in a comment
• Commit unit tests in branch fix-parts-of-speech-never-loader

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands and usage tips.}

[github-actions]

UI unit Tests

  1 files  ±0   50 suites  ±0   23s ⏱️ -2s
138 tests ±0  138 ✅ ±0  0 💤 ±0  0 ❌ ±0 
203 runs  ±0  203 ✅ ±0  0 💤 ±0  0 ❌ ±0 

Results for commit 4be66dd. ± Comparison against base commit a096927.

♻️ This comment has been updated with latest results.

[argos-ci]

The latest updates on your projects. Learn more about Argos notifications ↗︎

Build	Status	Details	Updated (UTC)
default (Inspect)	✅ No changes detected	-	Feb 10, 2026, 1:23 PM

[github-actions]

C# Unit Tests

146 tests  ±0   146 ✅ ±0   19s ⏱️ ±0s
 22 suites ±0     0 💤 ±0 
  1 files   ±0     0 ❌ ±0 

Results for commit 4be66dd. ± Comparison against base commit a096927.

[coderabbitai]

Actionable comments posted: 3

🤖 Fix all issues with AI agents

In `@frontend/viewer/src/project/project-context.svelte.ts`:
- Around line 119-122: The queued resource factories in setup()
(this.#queuedResources → factory(args.api)) get invoked twice because Svelte's
reactive resource source callback also calls factory(api); fix by ensuring only
one path runs: when you flush queued factories in setup(), mark those resources
as initialized (e.g., move their keys into a this.#initializedResources set or
delete them from this.#queuedResources) so the reactive resource source callback
can check that marker and early-return if the resource was already initialized;
alternatively, skip the flush and rely solely on the reactive path—apply the
change around the flush code that calls factory(args.api) and in the resource
source callback (the function that later calls factory(api)) to consult the same
marker before invoking factory.
- Around line 154-158: The queued factory promise can reject and currently
causes an unhandled rejection; update the callback stored in
this.#queuedResources[resourceId] (the function using
factory(api).then(res.mutate)) to handle errors by adding a .catch handler on
the Promise returned by factory(api) and surface the error (e.g., call a
resource error handler or log it) so failures don't become unhandled
rejections—ensure the callback uses factory(api).then(res.mutate).catch(err => {
/* handle or log err and notify resource */ }).
- Around line 120-121: The forEach callback currently uses the concise arrow
form which implicitly returns a value (Object.values(...).forEach(factory =>
factory(args.api));); update the callback to a block-bodied arrow that calls the
factory without returning anything—i.e., locate the Object.values<(api:
IMiniLcmJsInvokable) => void>(this.#queuedResources).forEach(...) line and
change the callback to use braces so it becomes factory => { factory(args.api);
} to satisfy the lint rule.