Fix #2663

[Nathoufresh]

All right after some testing I found out we just need to (setq history-add-new-input nil) and no modifications to ivy--update-history.

Ivy-read calls read-from-minibuffer which will add the selection to history if history-add-new-input is non-nil.

My issue is gone and the test are ok.

But now it will break some workflow like if someone prefer let's say find-file instead of counsel-find-file the history won't work. It's not really efficient but as a workaround instead of passing the real hist var to read-from-minibuffer we can pass a copy of it and so we can leave history-add-new-input as it is.

[Nathoufresh]

Not working with copy-sequence. I have no idea why

[basil-conto]

Not working with copy-sequence.

What isn't working exactly?

What about copy-alist?

[Nathoufresh]

Most tests fails with copy-squence or copy-alist but everything passe with copy-tree.
Logs.

From the copy-tree's docs

If TREE is a cons cell, this recursively copies both its car and its cdr.
Contrast to ‘copy-sequence’, which copies only along the cdrs.

[basil-conto]

Most tests fails with copy-squence or copy-alist

That's expected, because hist is usually a symbol, so passing it directly to copy-sequence or copy-alist is wrong.

but everything passe with copy-tree.

That's because copy-tree returns its argument unchanged if it is an atom.

That means this PR is a no-op, so I'm wondering which issue it's trying to solve, and whether it's an issue at all.

[Nathoufresh]

When you invoke counsel-M-x and you type ^ev buf and select eval-buffer in the history there will be eval-buffer then ^ev buf which I find not really convenient. An easy way to fix it is to set history-add-new-input to nil so you'd have only ^ev buf added to the history. But if let's say I don't use counsel for a command then I'll have no history at all for that command. I'm just trying to find a way to remove the eval-buffer while keeping history-add-new-input to true.

I think when I created this PR I've lost myself a bit and yeah, my method is definitely not working...

[basil-conto]

When you invoke counsel-M-x and you type ^ev buf and select eval-buffer in the history there will be eval-buffer then ^ev buf which I find not really convenient.

I agree, but I'm not sure whether @abo-abo introduced this intentionally.

I'm just trying to find a way to remove the eval-buffer while keeping history-add-new-input to true.

What if some users prefer the opposite: to keep eval-buffer in the history instead of ^ev, for consistency and better integration with vanilla Emacs completion? (In fact I think this discussion has come up before.)

So I think this behaviour should be customisable somehow.