This feature already exists, however you may have a faulty metacity configuration. Can you run:

$ gconftool-2 -u /apps/metacity/global_keybindings/cycle_windows_backward

and see if this fixes the problem?

I did that, but I still cannot use Alt+Shift+Tab.

Er, excuse me, sorry, the *correct* binding is switch_windows_backward:

$ gconftool-2 -u /apps/metacity/global_keybindings/switch_windows_backward

That does make it work.

Is this a broken remnant from my previous Gnome 2 config, so this will generally work for new users (in which case I am fine with closing the bug), or is there something outstanding that should still be done to leave the bug open?

(In reply to comment #4)
> That does make it work.
> 
> Is this a broken remnant from my previous Gnome 2 config, so this will
> generally work for new users (in which case I am fine with closing the bug), or
> is there something outstanding that should still be done to leave the bug open?

Yes, it's what you once configured for metacity (which should also have worked, whatever it was).

it seems that it's broken by default on ubuntu, for people who have used compiz, because switch_windows_backward is set to "Alt+Shift+Tab", which you would think would work, but it doesn't. We should probably fix that.

(In reply to comment #6)
> it seems that it's broken by default on ubuntu, for people who have used
> compiz, because switch_windows_backward is set to "Alt+Shift+Tab", which you
> would think would work, but it doesn't. We should probably fix that.

Indeed, I just tried and it does bring up the Alt+Tab popup but doesn't work past that.

Although the original reporter here isn't using Ubuntu, the Ubuntu problem is probably the same yes. For the record, here's the Ubuntu change that sets switch_windows_backward:

https://bugs.launchpad.net/ubuntu/+source/metacity/+bug/150702

My Gnome Shell/Mutter install is on Fedora, but the old configuration was from Ubuntu.

Same issue - Alt+Tab Works; Alt+Shift+Tab Failed. Setting /apps/metacity/global_keybindings/switch_windows_backward to disabled solved it.

This is coming from a fairly fresh Xubuntu 11.04 Install which then replaced with Gnome-Shell and XFCE removed. Gnome-Shell was installed through the GNOME3 PPA.

I ran into this issue on Fedora (#711251), possibly because it was a machine that was previously running Ubuntu. In any case, here's an interesting bit of conversation:

<magcius> The bug is that we ignore the switch_windows_backward key, which IIRC was on purpose for some reason
<danw> well, the bug is more that we *don't* ignore it
<danw> it was always the case that if the key was unset, then it was assumed to be switch_windows+Shift. it apparently used to be the case that you could set it explicitly to Alt+Shift+Tab if you wanted (but not anything else), but at some point it got broken to not allow even that
<magcius> https://bugzilla.gnome.org/show_bug.cgi?id=639341#c15

Created attachment 191369 [details] [review]
altTab: honor switch_*_backward key binding actions

Review of attachment 191369 [details] [review]:

You should also modify windowManager.js so that it works without an alt-tab popup open.

A better commit message could be better as well:

  We weren't honoring SWITCH_*_BACKWARD key bindings before, causing the
  unfortunate effect of making the switcher completely useless.

Created attachment 191370 [details] [review]
altTab: honor switch_*_backward key binding actions

This should fix it for good. From the discussion in bug 639341 I concluded, at the time, that these switch_*_backward key bindings were to be marked as deprecated and removed them. I didn't think of people upgrading nor that distributors (namely Ubuntu) were patching metacity to explicitly set those keys.

(In reply to comment #12)
> A better commit message could be better as well:
> 
>   We weren't honoring SWITCH_*_BACKWARD key bindings before, causing the
>   unfortunate effect of making the switcher completely useless.

Feel free to add one when committing.

Review of attachment 191370 [details] [review]:

::: js/ui/altTab.js
@@ +163,3 @@
             }
+        } else if (binding == 'switch_group_backward') {
+            this._select(0, this._appIcons[0].cachedWindows.length - 1);

Hm. I'm not sure I like this.

Can we try something like:

    // Additionally match "*_backward" bindings.
    if (/_backward$/.test(binding)) {
        backward = true;
        binding = binding.slice(0, -'_backward'.length);
    }

@@ +231,3 @@
             this._select(this._currentApp, backwards ? this._previousWindow() : this._nextWindow());
+        } else if (action == Meta.KeyBindingAction.SWITCH_GROUP_BACKWARD) {
+            this._select(this._currentApp, this._previousWindow());

Similarly here, it might be nicer to just set the backward flag:

    if (action == Meta.KeyBindingAction.SWITCH_GROUP_BACKWARD ||
        action == Meta.KeyBindingAction.SWITCH_WINDOWS_BACKWARD)
        backward = true;

Review of attachment 191370 [details] [review]:

::: js/ui/altTab.js
@@ +163,3 @@
             }
+        } else if (binding == 'switch_group_backward') {
+            this._select(0, this._appIcons[0].cachedWindows.length - 1);

That just doesn't "read" as nicely IMO.

@@ +231,3 @@
             this._select(this._currentApp, backwards ? this._previousWindow() : this._nextWindow());
+        } else if (action == Meta.KeyBindingAction.SWITCH_GROUP_BACKWARD) {
+            this._select(this._currentApp, this._previousWindow());

The backward flag here is really just indicating that Shift is pressed, so no, I don't think this would be better. 

Actually to be really complete it should be like this:

        } else if (action == Meta.KeyBindingAction.SWITCH_GROUP_BACKWARD) {
            this._select(this._currentApp, backwards ? this._nextWindow() : this._previousWindow());

Which is how metacity itself deals with this _BACKWARD bindings and Shift being an *additional* check on top of it. But this is just obfuscating the code without much real gain IMO. Who in their right mind is going to set SWITCH_GROUP_BACKWARD to, say <Alt>P and then expect <Alt><Shift>P to move the selector forward?

Big fail on my reply to you here Jasper :-) This is what I meant:

(In reply to comment #15)
> Hm. I'm not sure I like this.
> 
> Can we try something like:
> 
>     // Additionally match "*_backward" bindings.
>     if (/_backward$/.test(binding)) {
>         backward = true;
>         binding = binding.slice(0, -'_backward'.length);
>     }

That just doesn't "read" as nicely IMO.

The rest is understandable I guess.

Yeah, I'm happy with this version of the patch.

Rui, you should apply for a GNOME git account
(http://live.gnome.org/NewAccounts) so you can start committing your own
patches.

Attachment 191370 [details] pushed as 8db1ff8 - altTab: honor switch_*_backward key binding actions