Make undo-tree an optional dependency

[wasamasa]

I'd be thrilled if people tested this. Maybe even if a starter kit like Doom or Spacemacs enabled it. CC @hlissner @syl20bnr

[fbergroth]

Using undo-redo, I get an error when evil-redo calls (undo-redo nil), i.e. when count is nil. It seems to work if I provide a digit argument.

[wasamasa]

@fbergroth Thanks, I've fixed this now.

[leungbk]

Thanks for doing this. It's working well for me.

[leungbk]

Since goto-chg.el is still a dependency, and since undo-tree is a dependency of goto-chg.el, evil still winds up depending on undo-tree. From a quick skim of undo-tree's code, it doesn't look like the advice and hooks it sets are particularly intrusive when global-undo-tree-mode is turned off, though.

[wasamasa]

Hm, that's a good point. The code of goto-chg.el does have a soft dependency on undo-tree.el, but it still does declare a hard dependency on it and errors out if it cannot be loaded up successfully.

[TheBB]

Seems to work well for me, using the native Emacs (28+) functions.

[fbergroth]

Perhaps (interactive "*p") (error if buffer is read-only)?

[wasamasa]

Hm, it doesn't seem to ever hit that case sinde read-only buffers tend to not have any undo data in the first place or allowing you to edit them to create it.

[fbergroth]

Ah ok. A user could potentially enable read-only-mode after having made some changes though. I just noticed that undo-redo used "*p".

[wasamasa]

Fair enough, fixed.

[wasamasa]

I've found that setting this to undo-tree only works if you've made sure to evaluate (global-undo-tree-mode) first. For some reason I haven't been successful with enabling it automatically for users, but eh, I can live with that.

[TheBB]

Might be worth pointing that out somewhere, like the docstring of the evil-undo-system variable at least.

[wasamasa]

Alright, I've found out my issue and pushed a fix. Now you'll get a helpful error message if undo-tree-mode has not been enabled.

[wasamasa]

Before merging this I'd like to try removing the hard dependency of goto-chg.el on undo-tree.el, otherwise the changes to the README claiming that Evil no longer requires undo-tree.el aren't true.

Edit: emacs-evil/goto-chg@5c057c8

[wasamasa]

I went through the issue tracker and noticed that this PR might end up closing the following issues as well:

• Compile warning: undo-elt-crosses-region is an obsolete function #1275 (undo-tree is no longer vendored)
• undo-tree generating an error when starting Emacs #1144 (undo-tree is no longer the default)
• Remove undo-tree dependency #1074 (the issue that prompted this PR)
• Redo cursor position is broken. #601 (specific to undo-tree, cosmetic)
• When saving the buffer, change list got emptied. #522 (specific to undo-tree)
• undo-tree-undo doesn't work properly with a region selected #418 (specific to undo-tree, resolved by using undo-fu)

[TheBB]

Good work. Thanks!

[ColemanGariety]

I now get the error Customize 'evil-undo-system' for redo functionality. when redoing in evil-mode. Is this expected to break? What's the suggested way to get redo now?

[leungbk]

I now get the error Customize 'evil-undo-system' for redo functionality. when redoing in evil-mode. Is this expected to break? What's the suggested way to get redo now?

(setq evil-undo-system PREFERRED-UNDO-SYSTEM)

The options are 'undo-tree (extension), 'undo-fu (extension), and (Emacs 28 only) 'undo-redo.

[TheBB]

Yeah it's expected. Do as the message says and customize that variable.