* Continue to use Riot user data directory if it exists and the new
one doesn't, so users stay logged in.
* Move the auto-launcher over to new name.
* Look for keytar password under both names.
The docs (https://www.atom.pe/docs/api/sandbox-option/) say we should be using the browser-native `window.open` implementation, but in practice that appears very much false. Electron, no matter our set of options, appears to always make a hit to the ipcRenderer with `window.open` calls, causing the calling code to explode due to the sandbox making that impossible.
By using `app.enableSandbox()`, it puts the sandbox in place over all BrowserWindow objects, including the temporary ones which empirically are being created for `window.open`. We do not need to specify `sandbox: true` to the BrowserWindow with this approach, though uncommenting and therefore reintroducing the flag causes our lovely ipcRenderer error again.
As far as I can tell, the sandbox does actually get applied to the window though the fact that `sandbox: true` still does things despite the docs saying otherwise leaves me a bit uncomfortable.
Fixes https://github.com/vector-im/riot-web/issues/13719
This does two things related to search paths:
- also search for webapp symlink in the case of packaged applications
(e.g. Archlinux used to provide a symlink this way, linking to a
riot-web copy installed elsewhere, instead of creating an asar of it)
- also take dirname(asarPath) as a valid resources path, as this is how
riot-desktop currently is packaged. Fixesvector-im/riot-web#13190.
The react-sdk doesn't await the closing of the event index, this is done
so because of limitations in the lifecycle module. This wasn't a problem
since we used to just set the eventIndex variable to null. Nowadays we
wait for Tantivy to shutdown using a close() method on the index.
To avoid this being called multiple times while we're already closing the
index set the eventIndex variable to null before awaiting.
This fixes https://github.com/vector-im/riot-web/issues/12838.
A couple of changes to properties rather than functions and the
register*Protocol functionsd are now synchronous and so don't need
a completion callback.
