Sometimes we have specific code sections that can be run in parallel context by multiple threads with the obligation that some setup code must be executed exactly once before executing the actual functions. A simple solution is to just execute the existing setup function before the program enters a state from which parallel code can be executed from time to time.
The drawbacks of such an approach are the following ones:
- If the parallel function comes from a library, the user must not forget to call the setup function. That does not make the library easier to use.
- If the setup function is expensive in some way, and it might not even need to be executed in case the parallel functions that need this setup are not even always used, then we need code that decides when/if to run it.
In this recipe, we will have a look at std::call_once...
