Re: [PATCH 4/4] Make readline completers a little more useful

[Jose E. Marchesi]

Hi John.

Overall the patch looks good.

     struct pk_cmd file_cmd =
    -  {"file", "tf", "", 0, NULL, pk_cmd_file, "file (FILENAME|#ID)"};
    +  {"file", "tf", "", 0, NULL, pk_cmd_file, "file (FILENAME|#ID)", 
rl_filename_completion_function};

I like the idea of associating dot-commands with completers, but don't
you have to expand struct pk_cmd with a new pointer?  I can't find that
in any of the patches you sent...

    +extern struct pk_cmd *dot_cmds[];
    +extern struct pk_cmd null_cmd;
    +
    +static char *
    +poke_completion_function (const char *x, int state)
    +{
    +  static int idx = 0;
    +  static struct pkl_ast_node_iter iter;
    +  pkl_env env = pkl_get_env (poke_compiler);
    +  if (state == 0)
    +    {
    +      pkl_env_iter_begin (env, &iter);
    +      idx = 0;
    +    }
    +  else
    +    {
    +      if (pkl_env_iter_end (env, &iter))
    +   idx++;
    +      else
    +   pkl_env_iter_next (env, &iter);
    +    }
    +
    +  size_t len = strlen (x);
    +
    +  /* "Normal" commands.  */
    +  for (;;)
    +    {
    +      if (pkl_env_iter_end (env, &iter))
    +   break;
    +
    +      pkl_ast_node decl_name = PKL_AST_DECL_NAME (iter.node);
    +      const char *name = PKL_AST_IDENTIFIER_POINTER (decl_name);
    +      if (0 != strncmp (name, x, len))
    +   {
    +     pkl_env_iter_next (env, &iter);
    +          continue;
    +   }
    +      return  strdup (name);
    +    }
    +
    +  /* Dot commands */
    +  for (;;)
    +    {
    +      struct pk_cmd **c = dot_cmds + idx;
    +      if (*c == &null_cmd)
    +   break;
    +
    +      char *name = xmalloc (strlen ( (*c)->name) + 1);
    +      strcpy (name, ".");
    +      strcat (name, (*c)->name);
    +      if (0 !=  strncmp (name, x, len))
    +   {
    +     free (name);
    +     idx++;
    +     continue;
    +   }
    +      return name;
    +    }
    +
    +  return NULL;
    +}

What do you think about putting the completion logic for
compiler-dependent entities in the compiler?  poke_completion_function
would call pkl_complete instead of looping through the declarations
itself.

My main concern is that I expect for the logic to become more complex
(include certain declarations but no others, etc) and I hesitate to put
that logic in pk-repl.c.  My guts tell me the right place where to do
that is the compiler...