bug#64927: 30.0.50; kill-ring with no X

[Stefan Monnier]

>> The best we can do is to use `key-valid-p` as a best-effort test to
>> decide whether we're in the presence of the new or the old format, but
>> it will inevitably suffer from false positives/negatives.
>
> Not sure I understand: if key-valid-p returns nil, what do you suggest
> to do with "keys" such as those produced in the scenario of this bug
> report?

I suggest we do:

    diff --git a/src/keymap.c b/src/keymap.c
    index da2af98c2d6..eec3e884935 100644
    --- a/src/keymap.c
    +++ b/src/keymap.c
    @@ -1066,7 +1066,7 @@ possibly_translate_key_sequence (Lisp_Object key, 
ptrdiff_t *length)
                  build_string ("`key-valid-p' is not defined, so this syntax 
can't be used: %s"),
                  key);
           if (NILP (call1 (Qkey_valid_p, AREF (key, 0))))
    -        xsignal2 (Qerror, build_string ("Invalid `key-parse' syntax: %S"), 
key);
    +        return key;
           key = call1 (Qkey_parse, AREF (key, 0));
           *length = CHECK_VECTOR_OR_STRING (key);
           if (*length == 0)

With this change,

    (let ((map (make-sparse-keymap))) (define-key map ["he"] 'hello) map)

returns (keymap ("he" . hello)) as before, and emits a message:

    Key sequence contains invalid event he

whereas

    (let ((map (make-sparse-keymap))) (define-key map ["h e"] 'hello) map)

returns (keymap (104 keymap (101 . hello))).

I think this new [KBDSTRING] syntax was probably a mistake, in
retrospect: it made sense at the time, but we should have removed it
afterwards when we introduced the new `key-*` functions.


        Stefan