Bug 280684 – [Spiel] speech synthesis does not work with bubblewrap sandbox on

• Component: WebKitGTK
• Status: NEW

[q66]

Added in https://bugs.webkit.org/show_bug.cgi?id=270742.

The speech synthesis does not work because libspiel is unable to find any providers. This is because our xdg-dbus-proxy configuration does not allow the provider names to be seen in the filter. Unfortunately, we cannot wildcard them because xdg-dbus-proxy does not allow wildcarding prefixes. That clashes with how the speech provider names work; e.g. for espeak you have `org.espeak.Speech.Provider`, for something else you have `org.somethingelse.Speech.Provider` (the `.Speech.Provider` suffix is the same always). Since the dbus proxy does not filter them correctly, the `ListNames` and `ListActivatableNames` output does not include the provider (which libspiel checks for) and it goes undetected.

I was able to make it work specifically for espeak with this little patch:

```
--- a/Source/WebKit/UIProcess/Launcher/glib/XDGDBusProxy.cpp
+++ b/Source/WebKit/UIProcess/Launcher/glib/XDGDBusProxy.cpp
@@ -85,6 +85,10 @@ std::optional<CString> XDGDBusProxy::dbusSessionProxy(const char* baseDirectory,
     if (allowPortals == AllowPortals::Yes)
         m_args.append("--talk=org.freedesktop.portal.Desktop");
 
+#if USE(SPIEL)
+    m_args.append("--talk=org.espeak.Speech.Provider");
+#endif
+
     if (!g_strcmp0(g_getenv("WEBKIT_ENABLE_DBUS_PROXY_LOGGING"), "1"))
         m_args.append("--log");
 
```

This is, of course, not correct, as it does not match providers generically (otherwise i'd open a pull request). I'm not sure what the best way would be to match them generically. Thoughts?

[Eitan Isaacson]

There are plans for a speech synthesis portal, but no work has started on that in earnest. Here is an open issue for that: https://github.com/project-spiel/libspiel/issues/19

[q66]

well, i'd say a portal is a bit orthogonal to this issue, since this is related to webkit-internal sandbox and webkit works fine in an env without portals (the portal is more relevant to a flatpaked environment...)

so we'll probably still want to let the unportaled provider to be seen from within the sandbox (just like everything else)

[Patrick Griffis]

The Speech Provider API seems fairly reasonable to expose - https://project-spiel.org/dbus_iface.html

My main concern is just that the host service isn't hardened/sandboxed itself and could be exploited.