Add option to switch to undo-fu

[TheBB]

Just a draft branch that I'll be playing around with for a while. Opening this as a heads up.

Hopefully this can end up closing #1074, #418 and #522.

[edkolev]

Nice initiative!

A couple of notes:

• let me know if you need need me to dogfood this feature - I'd be happy to test
• supporting both undo-tree and undo-fu might be a stretch (in terms of implementation cost, and maintenance cost). I suggest supporting just one one them, i.e. no need for evil-undo-provider

[TheBB]

Yeah I'd love for others to test.

Wasn't sure yet whether I wanted to rip out undo-tree completely or what. If testing goes well we can take that into account. One problem is that undo-fu has very strict Emacs version requirement. I'll check with the author if that's just because it's untested on older versions.

[TheBB]

Also goto-chg depends on undo-tree but I think it's a soft dependency.

[TheBB]

There's been a new release to undo-tree, which might mitigate some of the problems people have been seeing. I'll leave this PR in limbo for a while.

[jixiuf]

I would got on emacs28

Symbol's function definition is void: undo-tree-save-history-from-hook

I think this hooks should be added after function is defined

(add-hook 'write-file-functions #'undo-tree-save-history-from-hook)
(add-hook 'kill-buffer-hook #'undo-tree-save-history-from-hook)
(add-hook 'find-file-hook #'undo-tree-load-history-from-hook)

can not reach the author of undo-tree

[aaronjensen]

@jixiuf he responds to email, which you can find on his website: http://www.dr-qubit.org/undo-tree.html

That said, I think he is unlikely to make any changes for Emacs28 unless they are totally innocuous for current emacs versions.

The error you're seeing is surprising... they're changing the semantics of #'symbol or add-hook is doing validation now...

[TheBB]

I already reported that to him. I got a fixed version by email but nothing on git or ELPA yet. If someone else wants to bump it be my guest!

…

On Fri, Jan 24, 2020, 05:46 Aaron Jensen ***@***.***> wrote: @jixiuf <https://github.com/jixiuf> he responds to email, which you can find on his website: http://www.dr-qubit.org/undo-tree.html That said, I think he is unlikely to make any changes for Emacs28 unless they are totally innocuous for current emacs versions. The error you're seeing is surprising... they're changing the semantics of #'symbol or add-hook is doing validation now... — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub <#1229?email_source=notifications&email_token=AAEXG3Z4U2GSN4XS6KMYXQ3Q7JXCXA5CNFSM4J6MMKRKYY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOEJZWWDQ#issuecomment-577989390>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAEXG34PD43GVSW7UVCLF3TQ7JXCXANCNFSM4J6MMKRA> .

[TheBB]

Undo-tree 0.7.3 is now out and it should have fixed the above mentioned issue.

[ideasman42]

Any update on this?

Even in the case undo-tree is has been fixed, it's a fairly large and complex package, it installs hooks even when it's disabled.
It would be nice if there was a way to fully disable undo-tree in evil mode.

(add-hook 'post-gc-hook #'undo-tree-post-gc nil)
;; --- snip --- 
(add-hook 'write-file-functions #'undo-tree-save-history-from-hook)
(add-hook 'kill-buffer-hook #'undo-tree-save-history-from-hook)
(add-hook 'find-file-hook #'undo-tree-load-history-from-hook)

---

Undo-fu by comparison:

• doesn't install any hooks/advice/timers.
• it's 213-SLOC compared to 2740-SLOC.
• no GC / undo-data manipulation.
• no extra mode/state needed It just exposes undo and redo which operate on Emacs undo information.

Since this PR, I've published undo-fu-session a companion package that can save/load undo data from disk. (to match undo-tree's functionality)

[TheBB]

No updates yet. I wrote this to try undo-fu but at that time there were still bugs. Since then undo-tree received a promising update, so I decided I could focus on other things for the moment.

I'll get around to it eventually, but of course if someone else wants to implement it I'd be happy to review and merge.

I'm pretty sure I actually do want the provider variable though. 😄

[TheBB]

Note: we also need to double check that goto-chg works fine when undo-tree isn't present. It's supposed to, but since evil for the moment requires undo-tree I'm not sure how much those code paths have been tested.

[ideasman42]

Testing goto-last-change and goto-last-change-reverse, they both seem to be working properly, although the navigation isn't linearized (it has the behavior of Emacs undo, not undo-only).

This seems more like a feature that could be supported in goto-chg.el than a bug.

[condy0919]

Any updates?

[TheBB]

If there's significant interest in an option for choosing undo provider please feel free to clean up the PR and resubmit. I don't have much time for evil at the moment. I've received no reports of bugs in the newer undo-tree versions so I don't consider it a priority.

[wasamasa]

@jixiuf @aaronjensen Do you have a minimal reproduction example for that change in hook functions? I've tried the following on Emacs built from master and failed to reproduce it with the following in a file:

(defun my-test () (message "Test"))
(add-hook 'post-command-hook 'my-test)

[TheBB]

Closed in favor of #1360