Re: [PATCH 3/3] Refactor atfork handlers

[Florian Weimer <fweimer at redhat dot com>]

On 02/07/2018 06:16 PM, Adhemerval Zanella wrote:
> +  for (size_t i = 0; i < fork_handler_list_size (&fork_handlers);)
> +    {
> +      /* dynarray remove maintains element order, so update index iff there is
> +	 no removal.  */
> +      if (fork_handler_list_at (&fork_handlers, i)->dso_handle == dso_handle)
> +        fork_handler_list_remove (&fork_handlers, i);
> +      else
> +        i++;
> +    }

I thought a bit more about this.  Doesn't this lead to cubic run-time as 
DSOs are unloaded (quadratic run-time locally here, multiplied by the 
outer loop for unloading the DSOs)?

I think fork_handler_list_remove is the wrong abstraction here. 
Something like std::remove_if would be better, which moves each array 
element at most once even if multiple elements are removed during the 
scan.  Writing this generically in C is probably not worth the effort, 
so perhaps open-code that here?

Thanks,
Florian