Hi Dmitry, Eli,

I read the discussion you pointed me to me by Dmitry in
http://lists.gnu.org/archive/html/emacs-devel/2016-01/msg01235.html.

Eli, If I understand your concerns there, then the first and third
patches I proposed shouldn't in any way interfere with your use of
*xref*-related facilities.  If anything, they should improve it.

The second patch does indeed bring the problem known as "the
disappearing *xref* problem", so I mended that with a very simple fourth
patch, described below.

So now, to summarize, there are 4 patches in total, that I reattach:

* 0001-Honor-window-switching-intents-in-xref-find-definiti.patch

This fixes the problem with the non-deterministic behaviour of selecting
a window for displaying cross-references, as well the problem where the
initial window/frame switching intent is lost.

* 0002-Quit-the-xref-window-if-user-decides-to-go-to-a-ref.patch

This makes the xref buffer non-obstrusive by quitting it on
xref-goto-xref. This is a change to a current behaviour that was
specifically requested by Eli (the "the disappearing *xref* problem").

* 0003-Allow-split-window-sensibly-to-split-threshold-in-fu.patch

This extends the exception granted by split-window-sensibly to
single-window frames whose dimensions are below those of splitting
thresholds to consider multi-window frames where all but one window is
dedicated.

In practice, it fixes the case where

  C-x 4 . xref-backend-definitions RET n

would surprisingly pop-up a new frame if the original frame was already
small to start with.

This fix to window.el appears very sound to me, but if it is not desired
for whatever reason, a more localized fix in xref.el is also possible.

* 0004-Don-t-quit-xref-window-on-RET-only-on-C-u-RET.patch

This fixes the "disappearing *xref* problem", by bringing back the
default behaviour of not quitting the *xref* window on RET, although
allowing for that if the user types C-u RET instead.

João