Re: [Sbcl-commits] master: Change depthoid of SEQUENCE from 2 to 1.

[Christophe R. <cs...@ca...>]

"Douglas Katzman" <sn...@us...> writes:

>     Change depthoid of SEQUENCE from 2 to 1.
>     
>     If this were the wrong thing to do, there should be test failures,
>     but there weren't, so it must be right.

Nice theory :-)

However: I think the reason for a depthoid of 2 (and for the failures
you see with stream) are related to the need to keep the classoids, type
system and PCL classes consistent.  In particular, sequence and stream
(and file-stream) are conceptually mixin classes as well as abstract
superclasses, and there needs to be room in the classoid layout-inherits
to stuff structure-object or standard-object.  Conventionally, the
allowed superclass list is

  (defclass foo (stream standard-object) ...)
  (defclass bar (sequence standard-object) ...)

so there needs to be a gap to allow standard-object to go in the
layout.  (But looking at them now I am confused about how
SB-PCL::SLOT-OBJECT is handled)

I'm not sure why putting sequence at depth 1 works.  Maybe my mission
for next month can be to find ways to expose any behaviour difference...

Cheers,

Christophe

> ---
>  src/code/class.lisp | 7 ++++++-
>  1 file changed, 6 insertions(+), 1 deletion(-)
>
> diff --git a/src/code/class.lisp b/src/code/class.lisp
> index b71bc2b..c52aedd 100644
> --- a/src/code/class.lisp
> +++ b/src/code/class.lisp
> @@ -967,7 +967,7 @@ between the ~A definition and the ~A definition"
>       (sequence
>        :translation (or cons (member nil) vector extended-sequence)
>        :state :read-only
> -      :depth 2)
> +      :depth 1)
>       (vector
>        :translation vector :codes (#.sb!vm:complex-vector-widetag)
>        :direct-superclasses (array sequence)
> @@ -1194,6 +1194,11 @@ between the ~A definition and the ~A definition"
>        :inherits (symbol list sequence)
>        :direct-superclasses (symbol list)
>        :prototype-form 'nil)
> +     ;; These last few are strange. STREAM has only T as an ancestor,
> +     ;; so you'd think it would be at depth 1. FILE- and STRING-STREAM
> +     ;; each have STREAM and T as ancestors, so you'd think they'd be at depth
> +     ;; 1 greater than STREAM, instead of 2 greater. But changing any of
> +     ;; these to the "obvious" value makes various type checks go wrong.
>       (stream
>        :state :read-only
>        :depth 2)
>
> -----------------------------------------------------------------------
>
>
> hooks/post-receive