Thanks, this works great! I felt like it would be easy but I didn't understand enough of the view code.

stop failing when trying to push an identical history state

Makes sense, this is required so we don't error on tig blame 529182c -- README.adoc Enter q Enter.

I found a similar use case for the diff view.

• tig show 529182c -- README.adoc
• :23
  • The cursor should be on the line after "The latest version is", that is, the deleted line containing tig-2.5.3
• Press b to enter the blame view.
  • What is a bit strange here is that the blame view shows the commit that
    added the line with tig-2.5.4. I expect it to show the commit that
    added the line with tig-2.5.3. This would be consistent with how we
    blame other deletions, so this might be an off-by-one error in tig show
• Press d to show the diff of commit tig-2.5.4, and :23 to go to the same line (deleted tig-2.5.3).
• Press b to blame the same line again but this time we actually get to the
  tig-2.5.3 commit.
• Press d to show the diff of commit tig-2.5.3. Now there is no nice way to get back to the diff of commit tig-2.5.4.
  • If I press q twice, I get back to the diff view which is still on the
    stack, but it shows commit tig-2.5.4. So I think we want to teach the diff
    view how to save/restore state, so :back or q could both go back to the earlier commit.

There is an issue when the diff view is the initial view. argv_env.commit is not set and the commit selection for blaming is done against HEAD instead of 529182c.

Here is a quick fix but I'm not completely sure it's the right way to do:

--- a/src/diff.c
+++ b/src/diff.c
@@ -34,6 +34,9 @@ diff_open(struct view *view, enum open_flags flags)
 	};
 	enum status_code code;
 
+	if (is_initial_view(view) && opt_rev_args)
+		string_copy_rev(argv_env.commit, opt_rev_args[0]);
+
 	diff_save_line(view, view->private, flags);
 
 	code = begin_update(view, NULL, diff_argv, flags);

I've openend #1135 with my attempt at a different fix. It's a bit hacky, I'll recheck it soon.

The other issue with the diff-view history would be solved if :view-close could restore the per-view history, like :back does for the blame view. The diff view doesn't have it's own history yet.
Currently, when in the blame view, a :view-blame will not push a view to the view stack, but only modify the blame view's history.
We could simplify this by always pushing a new blame-view on :view-blame. This would mean that every view on the view stack corresponds to (at most) one view state, which we can restore on :view-close. Of course this might make :view-close worse for the blame view.