bug#27718: 26.0.50; (cl-typep instance-ab 'class-a) gives wrong answer when compiled

[Stefan Monnier]

> The problem is that the `put' doesn't take effect until runtime, so the
> compiler inlines `cl-typep' incorrectly.  Using `function-put' here (and
> `function-get' in `cl-typep') solves it (requires also the patches for
> #27016[1]), although it seems a bit coincidental that `class-a' happens
> to be a function as well, so I'm not sure if this is the right thing.
> Stefan, any thoughts?

Hmm... I agree that using the same approach as that of bug#27016 is
right, but indeed `name` doesn't refer to a function but to the name of
a type (the property doesn't apply to the value stored in the
symbol-function cell and shouldn't follow alias definitions).

So maybe rather than hijack function-get/put in bug#27016, we should
introduce a new "definitional put" with a corresponding `get`, and then
use that here as well as in function-put/get.

Maybe we could simply change put&get directly without introducing
yet-another slightly different get/put?



        Stefan


> From 4ea34c9da04c8eabc754f2ca64ef4e2ac7c59cac Mon Sep 17 00:00:00 2001
> From: Noam Postavsky <npostavs@gmail.com>
> Date: Fri, 21 Jul 2017 07:59:21 -0400
> Subject: [PATCH v1] Fix (cl-typep instance-ab 'class-a) compilation
>  (Bug#27718)

> Using `function-put' instead of `put' for the `cl-deftype-satisfies'
> property lets the compiler know about it while compiling the file so
> that any calls to `cl-typep' will be inlined correctly.
> * lisp/emacs-lisp/eieio.el (defclass): Use `function-put' for the
> `cl-deftype-satisfies' property.
> * lisp/emacs-lisp/cl-macs.el (cl-typep): Also use `function-get' to
> check the 'cl-deftype-satisfies' property.
> ---
>  lisp/emacs-lisp/cl-macs.el | 7 +++++--
>  lisp/emacs-lisp/eieio.el   | 2 +-
>  2 files changed, 6 insertions(+), 3 deletions(-)

> diff --git a/lisp/emacs-lisp/cl-macs.el b/lisp/emacs-lisp/cl-macs.el
> index 451e6490d7..f5a4f99aad 100644
> --- a/lisp/emacs-lisp/cl-macs.el
> +++ b/lisp/emacs-lisp/cl-macs.el
> @@ -2991,8 +2991,11 @@ (define-inline cl-typep (val type)
>        ((and (pred symbolp) type (guard (get type 'cl-deftype-handler)))
>         (inline-quote
>          (cl-typep ,val ',(funcall (get type 'cl-deftype-handler)))))
> -      ((and (pred symbolp) type (guard (get type 'cl-deftype-satisfies)))
> -       (inline-quote (funcall #',(get type 'cl-deftype-satisfies) ,val)))
> +      ((and (pred symbolp) type (guard (or (function-get type 
> 'cl-deftype-satisfies)
> +                                           (get type 
> 'cl-deftype-satisfies))))
> +       (inline-quote (funcall #',(or (function-get type 
> 'cl-deftype-satisfies)
> +                                     (Get type 'cl-deftype-satisfies))
> +                              ,val)))
>        ((and (or 'nil 't) type) (inline-quote ',type))
>        ((and (pred symbolp) type)
>         (let* ((name (symbol-name type))
> diff --git a/lisp/emacs-lisp/eieio.el b/lisp/emacs-lisp/eieio.el
> index 1a7de55fce..2c333c16f4 100644
> --- a/lisp/emacs-lisp/eieio.el
> +++ b/lisp/emacs-lisp/eieio.el
> @@ -246,7 +246,7 @@ (defmacro defclass (name superclasses slots &rest 
> options-and-doc)
>         ;; test, so we can let typep have the CLOS documented behavior
>         ;; while keeping our above predicate clean.
 
> -       (put ',name 'cl-deftype-satisfies #',testsym2)
> +       (function-put ',name 'cl-deftype-satisfies #',testsym2)
 
>         (eieio-defclass-internal ',name ',superclasses ',slots 
> ',options-and-doc)
 
> -- 
> 2.11.1


>> You can see that tests 23 and 24 will fail when you do 'make
>> eieio-tests TEST_LOAD_EL=no'.

> This actually isn't true until the patch for #24402[2] is applied.

> [1]: https://debbugs.gnu.org/cgi/bugreport.cgi?bug=27016#140
> [2]: https://debbugs.gnu.org/cgi/bugreport.cgi?bug=24402#71