Re: [thread] upgrade_to_unique_lock takes an upgrade_lock, not an upgrade_mutex

From: Gavin Lambert (boost_at_[hidden])
Date: 2022-06-20 23:04:40

On 19/06/2022 07:52, Andrey Semashev wrote:
> I would pass a reference to upgrade_lock as an argument to every
> function that expects it to be locked and wants to upgrade it.
> 
>    void read_func()
>    {
>      upgrade_lock l(data_mut);
>      read(data);
>      if (foo) write_func(l);
>    }
> 
>    void write_func(upgrade_lock& l)
>    {
>      upgrade_to_unique_lock ll(l);
>      mutate(data);
>    }
> 
> This API is self-documenting, you won't be calling write_func without
> having an upgrade_lock.

Another option (which may be attractive if this is the only place that 
write_func is called, and is relatively small so does not benefit much 
from being a named method) is simply to inline it using an anonymous 
scope block:

     void read_func()
     {
         upgrade_lock l(data_mut);
         read(data);
         {
             upgrade_to_unique_lock ll(l);
             mutate(data);
         }
         read2(data); //back to shared+upgrade lock
     }