Re: [RFA 5/8] Allow defining a user command inside a user command

[Pedro Alves <palves at redhat dot com>]

On 04/19/2018 08:15 PM, Tom Tromey wrote:

>  build_command_line (enum command_control_type type, const char *args)
>  {
>    if ((args == NULL || *args == '\0')
> -      && (type == if_control || type == while_control))
> -    error (_("if/while commands require arguments."));
> +      && (type == if_control || type == while_control
> +	  || type == define_control))
> +    error (_("if/while/define commands require arguments."));

I'd vote for splitting those up:

    if (args == NULL || *args == '\0')
      {
        switch (type)
          {
          case if_control:
            error (_("if commands require arguments."));
          case while_control:
            error (_("while commands require arguments."));
          case define_control:
            error (_("define commands require arguments."));
          }
      }

> +/* Define a user-defined command.  If COMMANDS is NULL, then this is
> +   an interactive call and the commands will be read from the user.

Isn't this conflating top-level "define" command, with interactive
input?  I imagine that a top-level "define" command in e.g., gdbinit will
considered "interactive call" according to the above, while in truth,
it's not really interactive, according to FROM_TTY.

> +   Otherwise, it is a "define" command in a script and the commands
> +   are provided.  

Similarly, can't you write a "define" command inside a "define"
command interactively on the command line?

>In the non-interactive case, various prompts and
> +   warnings are disabled.  */
> +
>  static void
> -define_command (const char *comname, int from_tty)
> +do_define_command (const char *comname, int from_tty,
> +		   const counted_command_line *commands)
>  {

Thanks,
Pedro Alves