RE: lru-window, window--try-to-split-window, split-width-threshold

["Drew Adams" <drew.adams@oracle.com>]

>  > `split-width-threshold' was apparently introduced 
>  > recently. Depending on its value, the lru-window
>  > can now get split horizontally to display *Completions*
>  > during minibuffer input.
>  >
>  > I have code that changes the display of completions 
>  > (columns, spacing etc.), depending on the width
>  > *Completions* will have when it is displayed. Before it
>  > is displayed, I obtain (window-width (get-lru-window)) and 
>  > use that as the assumed window width.
>  >
>  > This is no longer adequate, however, because the effective 
>  > width might be, say, half of that if the lru window gets
>  > split horizontally to display *Completions*.
>  > And I can't simply use half of that width systematically, 
>  > because there might be no horizontal split, depending on
>  > `split-width-threshold'. I don't know a way to know ahead
>  > of time whether a horizontal split will occur (without
>  > reproducing all of the code/logic of
>  > `window--try-to-split-window' - checking for dedicated
>  > windows etc.).
> 
> So you can't fill the buffer _after_ the splitting occurred?  
> Relying on someone else to always choose the LRU window is
> not very clean in the first place.

Right you are.

I use code similar to that used in Emacs. The formatting is done inside
`with-output-to-temp-buffer', which displays the buffer only at the end.

>  > I tried this: (window--try-to-split-window (get-lru-window)),
>  > figuring that I might as  well split the lru-window myself,
>  > if it is going to be split, so that I can get its new width
>  > and use that.
>  >
>  > The problem with this is that once the split occurs the 
>  > window that was split is apparently no longer the lru window,
>  > so *Completions* gets displayed by splitting yet another window.
> 
> Are you sure?

No. ;-) In fact, looking again at the `display-buffer' code, I see what the
problem was:

(setq window-to-use
      (or (window--try-to-split-window
           (get-largest-window frame-to-use t))
          (window--try-to-split-window
           (get-lru-window frame-to-use t))))

I split the lru window, but then `display-buffer' split the largest window, so
two were split. And that's not all. To determine which window `display-buffer'
would use, I would in fact need to duplicate a good deal of the logic from
`display-buffer'.

So I took your suggestion and displayed *Completions* first. Thx.