Pane focus movement doesn't remember where it came from

[keithn]

Steps to reproduce

create a vertical split, then on the right pane, create a horizontal split
somewhat like this :

|-----------|
|    |  2   |
|  1 |______|
|    |   3  |
|____|______|

if you move focus from 1 to 2 by moveFocus right
then move focus from 2 to 3 by moveFocus down
then move focus from 3 to 1 by moveFocus left
then moveFocus right

Expected behavior

The focus should go back to 3. Otherwise toggling between 1 and 3 becomes a painful, especially if 2 is ends up more complicated with more panes.

Actual behavior

focus goes to 2

[moswald]

Agreed. Movement between panes should be MRU, rather than through move left/right/up/down which is ambiguous in a lot of cases.

[zadjii-msft]

Huh. This looks like it's the same thing as #2398, so I'm gonna clean this one out of the backlog. Thanks!

/dup #2398

[ghost]

Hi! We've identified this issue as a duplicate of another one that already exists on this Issue Tracker. This specific instance is being closed in favor of tracking the concern over on the referenced thread. Thanks for your report!

[mg-christian-axelsson]

I'd say there's a subtle, non-compatible difference between this and #2398. In this ticket the user want to remember focus in a multicolumn layout while in #2398 the user want to rely on visual alignment.

To clarify this ticket, if the layout was a bit more complicated like

|----------|
|    |  2  |
|    |_____|
|  1 |  3  |
|    |_____|
|    |  4  |
|____|_____|

And we moved from 3 and move left to 1 then if we move right again we should end up at 3. In #2398 if we rely upon visual alignment we should probably choose 2 or 4 (or maybe even 3).

In my opinion behavior in this ticket makes more sense than #2398 but comes down to user preference. In an ideal world perhaps we do both and let the user decide.

[zadjii-msft]

Ooooh. Okay. That's a great layout test case. Yea, I'll reopen this, to make sure we do this as well. Thanks!

IMO we should just do this always. In the "move right from 1" case, I don't think any one of 2, 3, or 4 is the obvious right one to pick. There's a good case for each of them. But moving to the most recent one - now that makes a lot of sense.

[mg-christian-axelsson]

Great to hear! There's of course a similar case with up/down when you have the tiling rotated 90 degrees. I haven't thought much about it and am not sure exactly how the two interact.

[Rosefield]

I would argue that this issue and #2398 are complimentary. #2398 defines how to find "neighbor" panes (visually) and this issue gives one possible strategy from selecting from those neighbor panes. Of course there is a dedicated command for move focus to the last used pane (which may not have existed when this issue was created.)

Consider the following case

-------------------------------
|       |         |           |
|   1   |    3    |    2      |
|       |---------|           |
|       |    4    |           |
-------------------------------

Where the tree is

  / \
 /\  2
1 /\
  3 4

Right now if you have 4 selected and move right, you'll hit 2. Then, if you move left

• current behavior selects 1 as focus
• 2398 will have you select either 3 or 4 since they are the two neighbors of 2, and I believe 3 would be selected since it is the "first-most" neighbor pane.
• this issue proposes to select from either 3 or 4, preferring 4 because it was more recent in history.

[zadjii-msft]

this issue proposes to select from either 3 or 4, preferring 4 because it was more recent in history.

I'd reckon that 4 is the correct pane to chose between the two. Both seem like reasonable choices, but I'd bet that it's more likely that someone wants to return to where they came from rather than just "the first one". Especially if we consider the #4692 (comment) example. Say the user moved from 4->1, then tried moving right again. They probably don't want to go to 4.

To get more complicated:

|----------------|
|  1  |    |  5  |
|_____|    |_____|
|  2  |  4 |  6  |
|_____|    |_____|
|  3  |    |  7  |
|_____|____|_____|

2->4->{right}->4->{left} should probably end on 2 again, since it was the MRU of [1, 2, 3].

That being said, we can probably do #2398, then do this in a follow-up when there's a "tie" between multiple panes.

[Rosefield]

If/when this gets implemented, I think it would be nice if it could be controlled by setting to enable/disable it.