Fix Svelte 5 reactive race when closing project settings modal

[sebastianhuus]

🧢 Changes

When closeSettings() sets modalProps to null, Svelte 5 can propagate updated props to child components before the outer {#if} guard destroys the block, causing TypeErrors like "null is not an object (evaluating '$.get(modalProps).state')" and "undefined is not an object (evaluating 'data.projectId')".

Introduce stableModalData — a $state updated only in $effect.pre when modalProps is non-null. Child components receive stableModalData (frozen at the last valid value during teardown) rather than reading modalProps directly, breaking the reactive propagation chain.

Related issues:

• Failed to remove a project #13252: A bit different but points to @mtsgrd who pushed related fixes a few days after the issue was raised. I think this compliments these and will help avoid similar problems.

☕️ Reasoning

Sent a little bug report in the Discord when I found I could no longer remove projects on v0.19.9. The previous approach was very minimal — each commit patched exactly the line that crashed in production telemetry:

5c4f013: telemetry showed null.close() → swap to handleModalClose
20e5a84: CI showed teardown race → add closeModal() with null guard
ac13c2e: error page showed "project not found" → reorder navigation/delete

If someone later adds a new modal type that closes itself via state mutation, it won't reintroduce this class of bug.

Media

Before fix, GitButler Client running 0.19.9

1. Remove project
2. Error toast shows

export-1777758137455.mp4

After fix:

1. Remove project
2. Project view is updated to a different project, project removed from project list and no error toast
   • Console complains about IPC cache issue, but as you can see the UI bug has been resolved.

export-1777758306650.mp4

[Byron]

Thanks a lot for taking a stab at this!

I can reproduce the issue, and it's a more recent one as it's first time I see this.

Screen.Recording.2026-05-03.at.16.36.08.mov

Previously it would fail after removing the project due to some other condition, but maybe with this PR all of these are fixed?

[sebastianhuus]

Yes hopefully so! I don't have anything else to add here so if there are no suggestions I'll send this to Copilot and we can hopefully get this merged?🤞

[Copilot]

Pull request overview

This PR hardens the desktop app's global modal router against a Svelte 5 teardown race when closing settings-related modals, so modal children do not briefly receive null/undefined modal data during unmount.

Changes:

• Add stableModalData, a last-known-good snapshot of the current global modal payload.
• Pass modal props/content and login-confirmation close handling through the stable snapshot instead of the live modal state.
• Tighten the modal render guard so the router only renders when both the current modal state and the stable snapshot are available.

[sebastianhuus]

Good spot but not really something I've introduced imo. Maybe plan this for a new PR since this is a narrow bug fix?

[sebastianhuus]

updated branch after new changes to main

[Copilot]

Pull request overview

Copilot reviewed 1 out of 1 changed files in this pull request and generated 2 comments.

[sebastianhuus]

I updated one comment which was inaccurate based on Copilot's feedback. I think rest is good.

There was one suggestion for adding a regression test, but I'm not sure how to set up a test to check for the target UI flow. If anyone wants to chime in and add that to the PR, feel free.

[Byron]

Thanks a lot!

I could reproduce that the original issue is fixed, and the project gets removed successfully.

For testing, one could probably add a test to playwright if it doesn't exist there yet, but it's nothing I personally would need to merge this.

Let's see if @estib-vega or @mtsgrd can merge the PR - I only looked at the outcome while thinking that the change is probably rather minimal.

---

More Notes

In my specific case, however, I ran into a couple of other, probably unrelated issues with recently removed projects still showing up in the project list after removal. But that state it was only in because it couldn't open a repo that was deleted on disk. Then the removal of that project wasn't offered anymore, also another issue.

Below there was a time when non-existing repos could be removed. Maybe previously there was a backend error code that the frontend would respond to that is gone now? Might be.