Re: display word wrapping

[Juanma Barranquero]

On 08 Jun 2004 08:38:00 +0200
David Kastrup <address@hidden> wrote:

> If an image type is known, I hope that it is present as just its
> name, and _not_ as a cons cell (jpeg . t) or so.
> 
> If not, we have a major incompatibility to the previous behavior.  It
> is bad enough already what happens if somebody decides to cycle
> through that variable just to display supported image types.

Yes, in my patch, image-types is an alist with (symbol . flag) items,
instead of a list of symbols.  It was suggested making it so on this
thread a few messages ago.

I'm not going to make a strong point on this; whatever is decided, I'll
implement.

However...

  - Kim *almost* suggested getting rid of image-types altogether,
    because image-types is an implementation detail.

  - No single .el, .c or .h file from the distribution uses image-types
    (other than image.c, and xdisp.c in which it is defined).

  - A Google search of "image-types" on gnu.emacs.* shows just one or
    two modules using image-types, in no particularly robust ways.

  - On delayed-loading environments (which could potentially be all, as
    delaying loading the libraries seems generally useful), image-types,
    even if a list as it was till now, can not be trusted to list
    supported image types, just supported types *already loaded*.

  - Delayed loading and image-types as it was till now, with the exact
    same semantics, are basically incompatible.  We could make:

    (defconst image-types '(jpeg tiff gif png postscript xpm pbm xbm))

    and use another variable as loaded-flags alist, but it's difficult
    to know what purpuse would it serve such image-types.  On Windows
    that wouldn't guarantee in any way that you can use png (for example),
    as libpng*.dll could be missing (that's the whole point of delayed
    loding).

  - You can easily fake an old style image-types with:

    (defvar my-image-types
     (let (result)
       (dolist (type '(xbm pbm xpm jpeg tiff gif png postcript) result)
         (when (image-type-available-p type)
           (push type result))))
     "Old style image-type.")

    and totally defeat delayed-loading.

> Can't we just keep the variable with the original meaning, and store
> the lookup cache data somewhere else?

We can keep the variable with a somewhat-similar meaning, but no with
the original one (unless we discard delayed-loading).  

                                                                Juanma