At first glance, undo seems like a simple thing expected of most software these days and hardly worth writing about. Indeed, when I say Emacs has a very powerful undo system—probably more so than any other text editor—you may wonder what could make an undo system powerful. So let’s start by considering two big problems most undo systems have:
- If you undo something, make some changes, then change your mind, what you undid is now lost and unrecoverable,
- If you make changes in two parts of the same file you cannot undo changes in the first part without undoing changes in the second part too.
Emacs comes with solutions to each of these out of the box. Read on to understand how it works and how we can improve upon the defaults even more.
Standard undo system
To deal with the first problem, it’s quite simple: Emacs stores undo commands themselves
in the undo history. To understand how this works, imagine a situation where you’ve made
two changes to a buffer and are now in state c. The history would look like this:
1a---b---c
2 ^
If you now undo twice, you will get back to state a, as you would expect, and the
history will look like this:
1a---b---c
2^
So far, so good, but what happens if we now make a non-undoing change such as entering
some new text to get into state b'. In most editors, states b and c would at this
point be lost, but in Emacs we get the following history:
1a---b---c---b---a---b'
2 ^
What’s happened is the moment a command breaks the chain of undos, the chain of undos
are themselves added to the undo history before any subsequent changes. This means you
can always get back to any previous state, including b and c.
This might sound quite hard to understand but, in fact, it’s actually quite intuitive and I used this standard undo system for many years.
Undo-tree
Another way to understand the states above is as a tree:
1 a
2 / \
3b b'
4| ^
5c
Now it’s perhaps possible to see that Emacs undo is actually doing a kind of tree traversal but, by default, you can’t see the tree, you just have to imagine it.
But what if it’s too difficult to imagine? That’s where undo-tree comes in. Undo-tree replaces the standard undo system with an alternative system that gives the standard undo/redo commands while still retaining full access to the tree when you need it. It comes with a graphical tree browser so you can view the undo tree and move anywhere within it.
I should have installed undo-tree years ago. As it happens, I’ve only started using it recently, but now an even better alternative is available.
Vundo
How I thought undo-tree worked was it used the standard Emacs undo system but merely enabled easier navigation through undo states by displaying a tree. This isn’t right, it actually replaces the undo system completely, but this is how vundo works. With vundo you use the standard undo system as described above, but you can display it as a tree and navigate through it when you need to.
But vundo would not be competitive with undo-tree if it weren’t for a couple of recent
changes to the standard Emacs undo system. These are the commands undo-only and
undo-redo. Unlike standard undo, undo-only will not undo undos and undo-redo
will only undo undos and not record itself as something to be undone. This might sound
a bit confusing, but you can think of undo-only and undo-redo as exposing just the
“normal” linear undo that most editors would provide.
I now have the following vundo config:
1(use-package vundo
2 :bind (("C-x u" . vundo)
3 ("C-/" . undo-only)
4 ("C-?" . undo-redo))
5 :config
6 (setq vundo-glyph-alist vundo-ascii-symbols))
To get persistent undo (ie. saving the undo history across Emacs sessions) there is undo-fu-session.
With this setup you get what undo-tree provided: the simple undo/redo system most of the time and access to the full tree when you need it. But because it uses the standard Emacs undo system it is simpler, potentially more robust and you get to use one of the most powerful Emacs undo features of all, as we will see next.
Undo in region
We’ve now covered problem number 1, but what about 2? A tragically little-known feature of the Emacs undo system is undo in region. Quite simply, if you select a region and undo, it will undo only within that region! How cool is that?
Undo-tree does support this, but it must be enabled by setting
undo-tree-enable-undo-in-region. However, it is known to be buggy and the undo-tree
author recommends against its use. But if we use vundo we can use it just fine.
Conclusion
The default Emacs undo system is the best there is. It’s one of the many small things that mean Emacs users never want to leave Emacs. Not only does it let you recover any previous state, you can even restrict your undoing to portions of the whole buffer.
But it wouldn’t really be Emacs if we didn’t still try to improve things. With just a couple of tweaks and a couple of extra packages we get an undo system that is easy to understand while losing none of its power and fully persistent between Emacs sessions.
Happy hacking!