Re: Convert a keyboard macro to equivalent Lisp code

[Juri Linkov <juri@jurta.org>]

>> To get useful results, some commands should provide their arguments
>> explicitly in the interactive spec instead of relying on global
>> variables.  So instead of useless `(self-insert-command 1)' typing `A'
>> will record `(self-insert-command 1 65)' with the following patch.
>> The same is for isearch.
>
> This will break all Elisp calls to self-insert-command (grep finds more
> than 100 of them in lisp/**/*.el) ;-(

I don't understand how it breaks Elisp calls.  The new second arg `last_char'
is optional.  When it's nil, this code is executed to preserve the
original behavior:

  if (NILP (last_char))
    last_char = last_command_event;

Or maybe not calling `translate_char' will break `self-insert-command'?

> So I think it's not an option.  Better would be to have your
> command-recording code provide hooks such that commands like
> self-insert-command can teach it how to turn them into Elisp code (in
> the case of self-insert-command it should probably use `insert').

Yes, where adding a new optional argument is impossible,
it's easy to record the right command.  For instance, if
`self-insert-command' needs to call `translate_char' to
record the right translated character, then after the call
to `translate_char' in `Fself_insert_command' we could replace
the recorded useless command `(self-insert-command 1)'
with the right command `(insert-char 1 65)':

=== modified file 'src/cmds.c'
--- src/cmds.c	2010-05-25 09:36:21 +0000
+++ src/cmds.c	2010-06-08 08:04:05 +0000
@@ -365,3 +370,3 @@
     int character = translate_char (Vtranslation_table_for_input,
				    XINT (last_command_event));

+    if (!NILP (current_kboard->defining_kbd_macro) && ! (minibuf_level > 0))
+      XSETCAR (current_kboard->Vlast_kbd_macro_command_history,
+	       Fcons (intern ("insert-char"),
+		      Fcons (make_number (character), Fcons (n, Qnil))));
+
     if (XINT (n) >= 2 && NILP (current_buffer->overwrite_mode))
       {
 	XSETFASTINT (n, XFASTINT (n) - 2);

> Of course, even better would be if the code run during macro recording
> is the code generated (so if the behavior is different from the normal
> command's behavior, you might see it during recording).

We could later add an option to verify to macro during recording.

-- 
Juri Linkov
http://www.jurta.org/emacs/