Jumping backward and forward in GNU Emacs
In my recent entry on writing Go with Emacs's lsp-mode, I noted that lsp-mode or more accurately lsp-ui has a 'peek' feature that winds up letting you jump to a definition or a reference of a thing, but I didn't know how to jump back to where you were before. The straightforward but limited answer to my question is that jumping back from a LSP peek is done with the M-, keybinding (which is surprisingly awkward to write about in text). This is not a special LSP key binding and function; instead it is a standard binding that runs xref-pop-marker-stack, which is part of GNU Emacs' standard xref package. This M-, binding is right next to the standard M-. and M-? xref bindings for jumping to definitions and references. It also works with go-mode's godef-jump function and its C-c C-j key binding.
(Lsp-ui doesn't set up any bindings for its 'peek' functions, but if you like what the 'peek' feature does in general you probably want to bind them to M-. and M-? in the lsp-ui-mode-map keybindings so that they take over from the xref versions. The xref versions still work in lsp-mode, it's just that they aren't as spiffy. This is convenient because it means that the standard xref binding 'C-x 4 .' can be used to immediately jump to a definition in another Emacs-level 'window'.)
I call this the limited answer for a couple of reasons. First, this only works in one direction; once you've jumped back, there is no general way to go forward again. You get to remember yourself what you did to jump forward and then do it again, which is easy if you jumped to a definition but not so straightforward if you jumped to a reference. Second, this isn't a general feature; it's specific to the xref package and to things that deliberately go out of their way to hook into it, which includes lsp-ui and go-mode. Because Emacs is ultimately a big ball of mud, any particular 'jump to thing' operation from any particular may or may not hook into the xref marker stack.
(A core Emacs concept is the mark, but core mark(s) are not directly tied to the xref marker stack. It's usually the case that things that use the xref marker stack will also push an entry onto the plain mark ring, but this is up to the whims of the package author. The plain mark ring is also context dependent on just what happened, with no universal 'jump back to where I was' operation. If you moved within a file you can return with C-u C-space, but if you moved to a different file you need to use C-x C-space instead. Using the wrong one gets bad results. M-, is universal in that it doesn't matter whether you moved within your current file or moved to another one, you always jump backward with the same key.)
The closest thing I've found in GNU Emacs to a browser style backwards and forwards navigation is a third party package called backward-forward (also gitlab). This specifically attempts to implement universal jumping in both directions, and it seems to work pretty well. Unfortunately its ring of navigation is global, not per (Emacs) window, but for my use this isn't fatal; I'm generally using Emacs within a single context anyway, rather than having several things at once the way I do in browsers.
Because I want browser style navigation, I've changed from the default backward-forward key bindings by removing its C-left and C-right bindings in favor of M-left and M-right (ie Alt-left and Alt-right, the standard browser key bindings for Back and Forward), and also added bindings for my mouse rocker buttons. How I have it set up so that it works on Fedora and Ubuntu 18.04 is as follows (using use-package, as everyone seems to these days):
(use-package backward-forward :demand :config (backward-forward-mode t) :bind (:map backward-forward-mode-map ("<C-left>" . nil) ("<C-right>" . nil) ("<M-left>" . backward-forward-previous-location) ("<M-right>" . backward-forward-next-location) ("<mouse-8>" . backward-forward-previous-location) ("<mouse-9>" . backward-forward-next-location) ) )
:demand is necessary on Ubuntu 18.04 to get
the key bindings to work. I don't know enough about Emacs to
PS: Normal Emacs and Lisp people would probably stack those stray )'s at the end of the last real line. One of my peculiarities in ELisp is that I don't; I would rather see a clear signal of where blocks end, rather than lose track of them in a stack of ')))'. Perhaps I will change this in time.
Comments on this page:Written on 08 September 2019.