mirror of
https://github.com/netbirdio/netbird.git
synced 2026-10-05 21:19:08 +02:00
Zoltan PappandClaude Opus 5
bc44cdc37a
* [client] Show the SSO login popup and open the browser from Go The browser-login popup was created hidden and relied on its own webview to size and show itself and to launch the external browser. On macOS a hidden WKWebView gets throttled or suspended (App Nap / hidden-window throttling), so on the first-use path nothing appeared and the browser never opened, leaving the session-expiration dialog disabled until the PKCE flow timed out. Reproduced by freezing the popup's WebContent process: the old code showed nothing, the new code shows the popup and opens the browser within 30 ms regardless of the webview state. Show and focus the popup from Go right after creation and launch the browser from Go on both the create and reuse paths. The popup's frontend no longer shows or focuses itself, so the browser keeps the foreground once it activates. This also fixes the reuse path, where a fragment-only SetURL kept the mounted React tree and the once-only guard skipped opening the browser for the new URI. Browser launch failures surface in the error dialog instead of being swallowed. * [client] Show every dialog window from Go once its frontend has painted Dialog windows (browser-login, session-expiration, install-progress, welcome, error) were created hidden and made visible only by their own webview's Show call after sizing. A hidden WKWebView on macOS can be throttled or suspended before that code runs, which left the window hidden forever. The main and settings windows already avoided this with the painted event plus a fallback timer, but that timer was armed on WindowRuntimeReady, which a frozen webview never reaches either. Route all dialogs through the same mechanism: the auto-size hook emits the painted event instead of showing the window, Go shows and focuses it on that event, and a fallback timer armed at creation shows it after 3 s regardless. The browser-login popup opens the browser in an after-show callback so the browser still lands in front of the popup, also on the fallback path. * [client] Tie install-progress hidden-window restore to the current popup CloseInstallProgress nils s.installProgress before calling w.Close(), so a replacement popup can open before the old window's WindowClosing event runs. The old callback then restored the windows the replacement had just hidden, because the restore sat outside the identity check. Guard the restore with the same check the state reset uses, and restore from CloseInstallProgress itself so the programmatic close path still re-shows the hidden windows — mirroring how CloseBrowserLogin already handles it. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> * [client] Correlate painted reports with the window generation that sent them A painted report carried only the window name, so a late report from a popup that was already closed and replaced marked its replacement ready. The replacement was then shown before its own frontend had rendered, which is the blank-dialog case this flow exists to prevent. Each dialog start URL now carries a monotonic generation token, echoed back by ReadySignal, and a report whose token no longer matches the live window is dropped. * [client] Separate a window being painted from its frontend being mounted One flag gated both showing a window and emitting to it, so the fallback timer set it for a frontend that had not subscribed yet: the queued events were flushed into a window that could not hear them, losing the login trigger and the settings tab selection. Showing is now gated on painted and emitting on mounted, and only a real frontend report sets mounted. The fallback timer also moved to its own helper so the runtime-ready hook can rearm it, giving the frontend a full budget to mount rather than sharing one with webview boot. * [client] Tag hidden windows with the popup that hid them Windows hidden while a popup owned the screen went into one untagged list, so whichever popup closed first restored all of them and emptied the list. An install started during SSO login re-showed the main window the login popup had deliberately hidden, and left the login popup with nothing to restore. Each entry now records the popup that hid it, and a restore releases only that popup's own entries. This also subsumes the manual filtering CloseRenewFlow did to keep its own session-expiration window from being re-shown. * [client] Cover the hidden-window bookkeeping with tests application.Window carries unexported methods, so the hide/restore paths could not be faked and the earlier tests could only assert which entries survived a restore, never which windows were actually shown. The bookkeeping now goes through hideableWindow, the four methods it needs, with the window enumeration and the main-window raise behind seams that are nil in production. That makes the case the owner tag exists for testable end to end: an install started during SSO login restores only the login popup it hid, and leaves the main window hidden until the login popup itself closes. * [client] Report the first paint from unstamped windows too The main and settings windows carry no generation token, so ReadySignal saw an empty generation that already matched the ref's initial value and never emitted the painted event. Those windows only became visible through the fallback timer, and their frontend was never marked mounted, so the login trigger and the requested settings tab stayed queued. Start the ref from null so the first report goes out regardless of the generation value. * [client] Hand covered windows over when a popup closes under another Closing the browser-login popup while the install-progress popup was still up restored the main window the login had hidden, even though the install popup was meant to own the screen until it finished. The owner tag on each hidden entry only stops a popup from restoring another's windows; it says nothing about what to do with its own when a second popup still covers them. Track which popups currently own the screen and, on restore, re-tag the entries another live popup covers to that popup instead of showing them. A popup is never handed its own window, so closing the popup on top still brings the one below back. --------- Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>