Re: [patch] [python] Prompt substitution

[Pedro Alves <pedro@codesourcery.com>]

On Wednesday 20 July 2011 12:33:24, Phil Muldoon wrote:
> @@ -258,7 +259,7 @@ void
>  display_gdb_prompt (char *new_prompt)
>  {
>    int prompt_length = 0;
> -  char *gdb_prompt = get_prompt ();
> +  char *actual_gdb_prompt = NULL;
>  
>    /* Reset the nesting depth used when trace-commands is set.  */
>    reset_command_nest_depth ();
> @@ -268,6 +269,25 @@ display_gdb_prompt (char *new_prompt)
>    if (!current_interp_display_prompt_p ())
>      return;
>  
> +  /* Get the prompt before the observers are called as observer hook
> +     functions may change the prompt.  Do not call observers on an
> +     explicit prompt change as passed to this function, as this forms
> +     a temporary prompt, IE, displayed but not set.  */
> +  if (! new_prompt)
> +    {
> +      char *post_gdb_prompt = NULL;
> +      char *pre_gdb_prompt = xstrdup (get_prompt ());
> +
> +      observer_notify_before_prompt (pre_gdb_prompt);
> +      post_gdb_prompt = get_prompt ();
> +
> +      /* If the observer changed the prompt, use that prompt.  */
> +      if (strcmp (pre_gdb_prompt, post_gdb_prompt) != 0)
> +       actual_gdb_prompt = post_gdb_prompt;
> +      
> +      xfree (pre_gdb_prompt);
> +    }
> +
>    if (sync_execution && is_running (inferior_ptid))
>      {
>        /* This is to trick readline into not trying to display the
> @@ -289,27 +309,35 @@ display_gdb_prompt (char *new_prompt)
>        return;
>      }
>  
> -  if (!new_prompt)
> +  /* If the observer changed the prompt, ACTUAL_GDB_PROMPT will not be
> +     NULL.  Otherwise, either copy the existing prompt, or set it to
> +     NEW_PROMPT.  */
> +  if (! actual_gdb_prompt)
>      {
> -      /* Just use the top of the prompt stack.  */
> -      prompt_length = strlen (PREFIX (0)) +
> -       strlen (SUFFIX (0)) +
> -       strlen (gdb_prompt) + 1;
> -
> -      new_prompt = (char *) alloca (prompt_length);
> -
> -      /* Prefix needs to have new line at end.  */
> -      strcpy (new_prompt, PREFIX (0));
> -      strcat (new_prompt, gdb_prompt);
> -      /* Suffix needs to have a new line at end and \032 \032 at
> -         beginning.  */
> -      strcat (new_prompt, SUFFIX (0));
> +      if (! new_prompt)
> +       {
> +         /* Just use the top of the prompt stack.  */
> +         prompt_length = strlen (PREFIX (0)) +
> +           strlen (SUFFIX (0)) +
> +           strlen (get_prompt()) + 1;
> +
> +         actual_gdb_prompt = (char *) alloca (prompt_length);
> +
> +         /* Prefix needs to have new line at end.  */
> +         strcpy (actual_gdb_prompt, PREFIX (0));
> +         strcat (actual_gdb_prompt, get_prompt());
> +         /* Suffix needs to have a new line at end and \032 \032 at
> +            beginning.  */
> +         strcat (actual_gdb_prompt, SUFFIX (0));
> +       }
> +      else
> +       actual_gdb_prompt = new_prompt;;
>      }

I believe this to be a bit broken in the annotations (prefix and
suffix != NULL) case.  If the observer changes the prompt, we won't print
either the prefix or the suffix, but will print directly what the
observer on the prompt itself only (get_prompt).

Make use of Matt's bug where the python prompt only works
the second time (so to trigger the strcmp):

$ gdb -quiet -ex 'set prompt (foo) ' -ex 'python def prompt(x): return "(bar) "' -ex 'python gdb.prompt_hook = prompt'
(foo) set annotate 2
(bar) 

While without the hook:

$ gdb -quiet
(gdb) set annotate 2

▒▒pre-prompt
(gdb) 
▒▒prompt

I believe it was not intended for the hook to be able
to override the prefix and the suffix as well.  Right?
I'm fixing this along with the get-rid-of-prompt-stack
change.

-- 
Pedro Alves