Re: feature request

[Dirk Herrmann]

On Sun, 20 May 2001, Han-Wen Nienhuys wrote:

> Since you're planning a new release, can I pretty-please ask you to
> remove the comments around
> 
> /* #define SCM_VOIDP_TEST    */
> 
> and change the definition for SCM from
> 
>     typedef void * SCM;
> 
> to
> 
>     typedef struct SCM_UNUSED_STRUCT { } * SCM;
> 
> 
> The latter catches more errors (eg. implicit conversions from
> character strings.)

I like your suggestion, however, it would be better if we could write
  typedef struct SCM_UNUSED_STRUCT * SCM;
(suggested by Marius) and never define SCM_UNUSED_STRUCT.  Thus, you would
even catch implicit conversions from "struct { }" types.  I have to admit,
though, that I don't know whether that would be legal ANSI C.

I just checked that guile compiles and links with
  typedef struct SCM_UNUSED_STRUCT * SCM;
without ever defining SCM_UNUSED_STRUCT, but it would be nice if someone
could confirm that this is valid ANSI C.

If so, I will apply that patch, if not so, I will apply Han-Wen's original
suggestion.  However, I intend to get rid of SCM_VOIDP_TEST and instead
distinguish between three possible values for SCM_DEBUG_TYPING_STRICTNESS:
  2: typedef union { struct { scm_bits_t n; } n; } SCM;
     (This one does not produce sensible code, but it catches _all_
      type mismatches when using SCM.)
  1: typedef struct scm_unused_struct * SCM;
     (This or Han-Wen's original suggestion will be the default.  It
      catches most type mismatches, but does not compromise code
      efficiency on any sane compiler.)
  0: typedef scm_bits_t SCM;
     (People having difficulties to get guile compiled with one of the
      above values, or which encounter performance drawbacks can use this 
      setting.)

Best regards,
Dirk Herrmann