Insertion of tabs shifts prior changes

Question

Insertion of tabs shifts prior changes

axelf4 opened this issue 3 years ago · 3 comments

axelf4 commented 3 years ago

After inserting a bunch of tabs going to the earlier changes puts the point in the wrong spots.

I am using the following packages:

evil: at commit emacs-evil/evil@17c635f
undo-tree: at commit with Timestamp: 2020-11-28 21:05:29
goto-chg: at commit 3ce1389
(plus a bunch of miscellaneous packages that hopefully are not to blame.)

To reproduce:

Set glc-default-span to 1.
Enter a new buffer.
Insert foo with ifoo<Esc>
Insert \n...\nbar with o<RET><RET><RET>bar<Esc>
Goto first line and prepend a bunch of Tab:s with ggI^v<Tab>^v<Tab>^v<Tab>^v<Tab>^v<Tab>^v<Tab>
The buffer now looks like
```
						foo


bar
```
Executing M-x goto-last-change now puts the cursor in the expected spot:
```
						█oo


bar
```
where █ denotes the cursor.
However, executing M-x goto-last-change again puts it in the wrong place:
```
						foo
█

bar
```
instead I would have expected it to put the cursor on the r, which is what I get if instead of Tab:s I inserted e.g. spaces.

Answer 1 · 2021-10-28T16:44:49.000Z

I cannot reproduce this, neither with my customized Emacs nor by a minimal session (run by executing make emacs from a Git checkout of Evil, then visiting goto-chg.el, evaluating its buffer, then following the repro instructions). I advise you to do the same before opening bug reports.

Answer 2 · 2021-10-28T17:23:28.000Z

I apologize for wasting your time. However I am able to reproduce it using make emacs still (asciinema), were you perhaps not enabling undo-tree? The precise steps:

Run make emacs in the Evil tree checked out at emacs-evil/evil@215c39d (master is buggy for me?)
Source undo-tree.el and goto-chg.el

Source

(setq glc-default-span 1)
(setq-default tab-width 4)
(add-hook 'evil-local-mode-hook #'undo-tree-mode)

Proceed with the steps outlined in the original issue

Answer 3 · 2021-10-28T21:41:31.000Z

You're right, I forgot about undo-tree and generally prefer not using it if possible. If it's enabled, goto-chg uses a different strategy for discovering the last change by using buffer-undo-tree instead of buffer-undo-list. It's best-effort code lifted from https://github.com/martinp26/goto-chg. The code is absolutely inscrutable for me and requires someone actually understanding undo-tree to fix it.