[v3] SingleInstance: os.Exit inside application.New skips the caller's defers (silent cleanup loss)

[midagedev]

Wails version family

v3 (alpha/beta)

Exact Wails version

v3.0.0-beta.15 (sites verified identical at master 9d49e8e and at beta.9/beta.12)

Operating System

All (the os.Exit call is in shared code; darwin/linux have one extra exit site each)

Description

When SingleInstance detects an already-running instance, application.New calls os.Exit directly:

• v3/pkg/application/application.go:206-218 — on alreadyRunningError: notifyFirstInstance(), then os.Exit(appOptions.SingleInstance.ExitCode)
• plus single_instance_darwin.go:82 and single_instance_linux.go:99

os.Exit terminates without running deferred functions (Go spec), so any cleanup the caller scheduled before calling New() — flushing a write-ahead file, releasing a lock file, closing a log — is silently skipped in the second process. New() looks like an ordinary constructor, so having opened resources before it is natural; the constraint is invisible in the API.

In our production app (gadak) the second-instance exit was skipping a persistence flush scheduled with defer. We had to reorder initialization so nothing requiring cleanup exists before New(), and pin that ordering with a test — that works, but every SingleInstance adopter has to discover it the same way.

What this issue is not asking: SingleInstance.ExitCode already covers the exit value (#3723), and the second instance correctly must not continue into Run() and show a window — no change to that. The ask is a seam for cleanup, in either shape:

1. New() returns a distinguished error (e.g. errors.Is(err, application.ErrAlreadyRunning)) after notifyFirstInstance(), and the caller exits — possibly opt-in to avoid breaking apps that don't check the error; or
2. a hook, e.g. SingleInstance.OnSecondInstanceExit func(), invoked after notifyFirstInstance() and before os.Exit.

Related: #3301 ("in idiomatic Go we return an error"); #4066 already converted service-startup fatals into returned errors; #5170 recently touched this path. Happy to send a PR for whichever shape you prefer.

To Reproduce

1. SingleInstance enabled; in main(), open any resource needing cleanup and schedule it with defer (e.g. defer f.Sync()), then call application.New(...).
2. Start the app once, then start a second instance.
3. The second process exits with ExitCode — the deferred cleanup never runs (observable with a defer that writes a sentinel file: the file is not written).

Expected behaviour

A way for the second instance to run its cleanup before terminating — via a returned error or a pre-exit hook — while wails keeps owning the "second instance does not open a window" behavior.

Attempted Fixes

Reordered our initialization so that nothing requiring cleanup exists before New(), pinned with a test. Works, but is a per-app rediscovery of an API-invisible constraint.