The call notation of CHARSET_INFO routines has some disadvantages:

• it looks cumbersome
• it exposes internal structure of CHARSET_INFO to the caller

Examples:

  // An example from storage/innodb/

  tmp_length = charset->coll->strnxfrm(charset, str, str_length,

                                       str_length, tmp_str,

                                       tmp_length, 0);

  // An example from storage/innodb/

  mbl = cs->cset->ctype(cs, &ctype, (uchar*) doc, (uchar*) end);

  // An example from storage/myisam/

  mbl= cs->cset->ctype(cs, &ctype, (uchar*)doc, (uchar*)end);

  // An example from storage/myisam/

  keyseg->charset->cset->fill(keyseg->charset,

                              (char*) pos + length,

                              keyseg->length - length,

                              ' ');

To make the call notation simple and proof to changes in CHARSET_INFO, lets do the following:

• Add pure C wrappers for all virtual functions in MY_CHARSET_HANDLER and MY_COLLATION_HANDLER, e.g.

  static inline void

  my_ci_fill(CHARSET_INFO *cs, char *to, size_t len, int ch)

  {

  (cs->cset->fill)(cs, to, len, ch);

  }

  Let's call all new functions using the my_ci_ prefix, to make it clear that the first argument is CHARSET_INFO.

• Add C++ methods into struct charset_info_st, like this:

  struct charset_info_st

  {

  #ifdef __cplusplus

  ...

  void fill(char *to, size_t len, int ch) const

  {

  (cset->fill)(this, to, len, ch);

  }

  ...

  size_t strnxfrm(uchar *dst, size_t dstlen, uint nweights,

  const uchar *src, size_t srclen, uint flags) const

  {

  return (coll->strnxfrm)(this,

  dst, dstlen, nweights,

  src, srclen, flags);

  }

  ...

  #endif

  };

so the code in the above examples will turn into:

  // C++ code

  tmp_length = charset->strnxfrm(str, str_length,

                                 str_length, tmp_str,

                                 tmp_length, 0);

  // C++ code

  mbl = cs->ctype(&ctype, (uchar*) doc, (uchar*) end);

  /* Pure C code */

  mbl= my_ci_ctype(cs, &ctype, (uchar*)doc, (uchar*)end);

  /* Pure C code */

  my_ci_fill(keyseg->charset, (char*) pos + length,

                              keyseg->length - length,

                              ' ');

The new notation is better, as it does not contain sequences like cs->cset-> and cs->coll->, and the CHARSET_INFO parameter is mentioned only one time (instead of two times), so the new style of the caller code:

• is shorter
• is less bug prone
• is future proof: it won't change if we change the structure CHARSET_INFO, e.g. decompose CHARSET_INFO into smaller pieces responsible for character set and collation properties. Only the wrapper functions and methods will change, the caller code will remain the same.