On 12/11/2012 09:42 PM, Pedro Alves wrote:
> On 12/11/2012 03:36 PM, ali_anwar wrote:
>> +  if (thread_count ())
>> +    {
>> +      struct thread_info *tp_array;
>> +      struct thread_info *tp;
>> +      int i, k;
>> +
>> +      /* Save a copy of the thread_list in case we execute detach
>> +         command.  */
>> +      tp_array =  xmalloc (sizeof (struct thread_info) * thread_count ());
>
> No need to compute the thread count twice, you can cache it.  No need to
> copy the whole thread structure.  Make this an array of a thread
> pointers, and then,
>
>> +      for (i = 0, tp = thread_list; tp; i++, tp = tp->next)
>> +        tp_array[i] = *tp;
>
>        ALL_THREADS (tp)
>          {
>            tp_array[i] = tp;
>            tp->refcount++;
>          }
>
> This increments the refcount of each current thread, so that attempts to
> delete it just mark it as deleted (so the C object remains valid).
>
>> +
>> +      for (k = 0; k != i; k++)
>> +        if (thread_alive (&tp_array[k]))
>
> and then write:
>
>        for (k = 0; k != i; k++)
>          {
>            if (thread_alive (tp_array[k]))
>              {
>                switch_to_thread (tp_array[k]->ptid);
>
>                printf_filtered (_("\nThread %d (%s):\n"),
>                                 (tp_array->num, target_pid_to_str (inferior_ptid));
>                execute_command (cmd, from_tty);
>                strcpy (cmd, saved_cmd);        /* Restore exact command used
>                                                   previously.  */
>              }
>          }
>
> And put this in a cleanup:
>
>        for (k = 0; k != i; k++)
>          tp_array[k]->refcount--;
>
> So that if the command throws an error, we still leave with the correct
> refcounts.
>
> The advantages are:
>
>   - less memory necessary for the array.
>   - handles the corner case of the target reusing a ptid (see
>     add_thread_silent).  IOW, this way, even if the command happens to
>     make the target reuse a ptid, "thread apply all" won't run the command
>     on that new threads my mistake.
>
>

I have tried to implement what you suggested in the attach patch. Does 
it look reasonable?

Thanks,
-Ali