Class LogSink (2.36.0)

We want to avoid synchronization overhead when checking if a log message is enabled. Most of the time, most messages will be disabled, so incurring the locking overhead on each message would be too expensive and would discourage developers from creating logs. Furthermore, missing a few messages while the change of state "propagates" to other threads does not affect the correctness of the program.

Note that memory_order_relaxed does not provide a compiler barrier either, so in theory stores into the atomic could be reordered by the optimizer. We have no reason to worry about that because all the writes are done inside a critical section protected by a mutex. The compiler cannot (or should not) reorder operations around those.

We want to avoid synchronization overhead when checking if a log message is enabled. Most of the time, most messages will be disabled, so incurring the locking overhead on each message would be too expensive and would discourage developers from creating logs. Furthermore, missing a few messages while the change of state "propagates" to other threads does not affect the correctness of the program.

Note that memory_order_relaxed does not provide a compiler barrier either, so in theory stores into the atomic could be reordered by the optimizer. We have no reason to worry about that because all the writes are done inside a critical section protected by a mutex. The compiler cannot (or should not) reorder operations around those.

This is also enabled if the "GOOGLE_CLOUD_CPP_ENABLE_CLOG" environment variable is set.

Note that this will remove the default logging backend.