WiFi password prompt empty when reconnecting

[Michał Sawicz]

This might be on purpose (so we don't leak the password on a locked phone), but compared to unity7 the WiFi password prompt does not come pre-filled with the password.

This makes it very annoying when you have a flaky WiFi connection, OTOH you can cancel the prompt and click on the network in the indicator and have NM try and reconnect with the previously set password.

Filing this so we can have a record of why it is like that.

ProblemType: Bug
DistroRelease: Ubuntu 15.04
Package: indicator-network 0.7.1+15.04.20160511-0ubuntu1
Uname: Linux 3.10.100-user-01769-g42b9fe3 aarch64
ApportVersion: 2.17.2-0ubuntu1.3touch1
Architecture: armhf
Date: Fri Jun 24 11:35:45 2016
InstallationDate: Installed on 2016-06-24 (0 days ago)
InstallationMedia: Ubuntu 15.04 - armhf (20160624-085657)
SourcePackage: indicator-network
UpgradeStatus: No upgrade log present (probably fresh install)
indicator-network.log:
 void nmofono::wifi::WifiLinkImpl::Private::ap_removed(const QDBusObjectPath&) : Tried to remove access point "/org/freedesktop/NetworkManager/AccessPoint/9" that has not been added.

 (process:5074): GLib-GIO-WARNING **: Error releasing name com.canonical.indicator.network: Połączenie jest zamknięte
 void nmofono::wifi::WifiLinkImpl::Private::ap_removed(const QDBusObjectPath&) : Tried to remove access point "/org/freedesktop/NetworkManager/AccessPoint/11" that has not been added.
 void nmofono::wifi::WifiLinkImpl::Private::ap_removed(const QDBusObjectPath&) : Tried to remove access point "/org/freedesktop/NetworkManager/AccessPoint/12" that has not been added.
upstart.indicator-network-secret-agent.log:
 QVariantDictMap agent::SecretAgent::GetSecrets(const QVariantDictMap&, const QDBusObjectPath&, const QString&, const QStringList&, uint) "/org/freedesktop/NetworkManager/Settings/11" "802-11-wireless-security" () 3
 QVariantDictMap agent::SecretAgent::GetSecrets(const QVariantDictMap&, const QDBusObjectPath&, const QString&, const QStringList&, uint) Requesting secret from user
 virtual notify::Notification::~Notification() Closing notification: 1
 virtual notify::Notification::~Notification() "Client ':1.68' tried to close notification 1, which it does not own or does not exist."
upstart.indicator-network.log:
 void nmofono::wifi::WifiLinkImpl::Private::ap_removed(const QDBusObjectPath&) : Tried to remove access point "/org/freedesktop/NetworkManager/AccessPoint/9" that has not been added.

 (process:5074): GLib-GIO-WARNING **: Error releasing name com.canonical.indicator.network: Połączenie jest zamknięte
 void nmofono::wifi::WifiLinkImpl::Private::ap_removed(const QDBusObjectPath&) : Tried to remove access point "/org/freedesktop/NetworkManager/AccessPoint/11" that has not been added.
 void nmofono::wifi::WifiLinkImpl::Private::ap_removed(const QDBusObjectPath&) : Tried to remove access point "/org/freedesktop/NetworkManager/AccessPoint/12" that has not been added.

[Matthew Paul Thomas (mpt)]

I need to fix this in the spec: I didn’t mention it in the text, but I included two wireframes of the situation, one showing the password field empty, the other pre-filled. Oops. <https://wiki.ubuntu.com/Networking#wi-fi-connecting-dialog>

That a device might be locked shouldn’t be a reason to blank the password field; we could omit the “Show password” checkbox in that case, or suppress the dialog altogether.

Unfortunately, probably ~70% of the time you see this dialog the password is correct, and you’re just experiencing bug 615239. Since that bug has existed since 2009 if not earlier, it may not be fixed soon, and if it won’t be, we should pre-fill the field.

If/when that bug is fixed, though, the equation becomes much murkier. We should pre-fill the field *if* the time it would save, in correcting typos, would be greater than the overall time taken to fix non-typo errors. I think we can assume that you would always need to reveal the password to tell that the error is a typo, but you might not need to reveal the password to tell/remember that it isn’t.

If so, where
P(typo) = probability that the password error is a typo;
P(non-typo) = probability that the password is completely wrong (e.g. outdated);
P(reveal) = probability that you’ll check “Show password” to see which of those two things it is;
t(reveal) = time taken to check “Show password”;
t(correct) = average time taken to correct a typo;
t(retype) = average time taken to retype the password from scratch instead;
P(unreplacing) = probability that you don’t know/remember that typing replaces selection; and
t(backspace) = time taken to tap Backspace instead of typing to replace the selection;

then we should pre-fill the field if
P(reveal) × P(typo) × (t(retype) – (t(reveal) + t(correct)))
>
P(reveal) × t(reveal) + P(non-typo) × P(unreplacing) × t(backspace).

Probably I’ve made at least one mistake there.