Ctrl+U Alt+A (C-u M-x) eats the first character

[pdg137]

To reproduce, type Ctrl+U Alt+A shell, and you will see "C-u M-x hell" in the minibuffer (with no "s").

I verified this with a fresh emacs with ergoemacs-mode-20131210.815 installed via Melpa, executing:

(package-initialize)
(setq ergoemacs-theme nil)
(setq ergoemacs-keyboard-layout "us")
(ergoemacs-mode 1)

[mattfidler]

What a devilish bug for a sunday...I will look into it.

[mattfidler]

I can't exactly reproduce it. However, Ctrl+u Alt+a doesn't even bring up an execute kbd-macro for me... Something is wrong.

[mattfidler]

I'm kind of stumped on this bug. I can't see where a read-key or unread-command-events wasn't handled properly. I even made sure that execute-extended-command is not called to ergoemacs-shortcut and that still doesn't work.

[pdg137]

Thanks for looking into it! I don't remember this being a problem with Xah Lee's package from before it was on MELPA. Have you considered doing something like a binary search to determine when it broke? I also wonder, since you are not seeing the same symptoms as I am, whether you have something like helm installed that could affect the behavior.

[pdg137]

I tried using "git bisect" to investigate further and found that the problem appears (for me, at least) at commit 95fa49c.

[mattfidler]

I now reproduce the behavior. I'm not sure why I wasn't reproducing it before. It does the same thing in helm and smex, unfortunately.

I tried reverting to the commit you pointed out and it didn't really work for me. I'm going to have to look into this more...

[mattfidler]

Thanks for the git bisect tip. Facinating.

[mattfidler]

By the way, an easy way to start a clean ergoemacs session is to use the key <menu> h z or C-h z.

[mattfidler]

You look like someone I know... Is your first name Tom?

[mattfidler]

I got to the same place....

[pdg137]

Sorry, it's Paul. Anyway, 95fa49c is the first bad commit; the parent, i.e. 95fa49c^ should work. It looks like the commit throws away a lot of keyboard bindings in a strange way, but I don't understand it because it is in the middle of a monster function. The code seems to have been cleaned up a lot since then, and the particular condition-case that broke Ctrl+U is no longer present.

The bad commit also caused some tests in ergoemacs-test to fail, but there don't seem to be tests for the bug that was supposedly fixed, and I don't know how write tests myself, so I have no idea about how to proceed.

[mattfidler]

Sometimes I can't figure out how to write a test for a bug. This is one that I can't figure out how to write a test for.

[mattfidler]

The funny thing about this bug, is ergoemacs-mode shouldn't be even calling this function.

By setting

(setq ergoemacs-show-true-bindings t)

The actual function bound to M-a is execute-extended-command, whereas bindings such as M-k run this function.

[mattfidler]

Even taking out the post-command-hook and pre-command-hook functions, this doesn't work.

[mattfidler]

By taking out all refenreces to this function, the bug still occurs. This is done by:

(defun ergoemacs-setup-keys-for-layout (layout &optional base-layout)
  "Setup keys based on a particular LAYOUT. All the keys are based on QWERTY layout."
  (ergoemacs-setup-translation layout base-layout)
  ;; Reset shortcuts layer.
  (setq ergoemacs-command-shortcuts-hash (make-hash-table :test 'equal))
  (let ((setup-ergoemacs-keymap nil))
    (ergoemacs-setup-keys-for-keymap ergoemacs-keymap))
...

[mattfidler]

I don't think that function is causing the bug....

[mattfidler]

Right now, I have a work-around. If execute-extended-command helm-M-x or smex is called pretend you typed a p. Then the p is eaten instead of the next key... Its a horrible work-around, but I think it solves the issue tempoaraily.

[pdg137]

Thanks for working so hard on this! By the way, the bug also occurs with some other prefixed commands like Ctrl+U Ctrl+O (though I'm not sure why one would use that.) When I was testing on commit 95fa49c, the function was definitely called many times after pressing Ctrl+U, and it reached condition-case but never add-to-list.

[mattfidler]

I think this bug occurs whenever there is a prefix argument and the command enters the minibuffer.

[mattfidler]

Hm. It may be the universal-argument-other-key that is eating the character.

(defun universal-argument-other-key (arg)
  (interactive "P")
  (setq prefix-arg arg)
  (let* ((key (this-command-keys))
     (keylist (listify-key-sequence key)))
    (setq unread-command-events
      (append (nthcdr universal-argument-num-events keylist)
          unread-command-events)))
  (reset-this-command-lengths)
  (restore-overriding-map))

[mattfidler]

It is run right after ergoemacs starts the minibuffer.

[mattfidler]

This should be fixed. Even for the Ctrl+o example

[pdg137]

The current version in MELPA now works great. Thanks!