When I first discovered Java's synchronized
blocks, I was intrigued. It's
pretty easy to forget to unlock a mutex, though RAII and unique_lock aid with
this, and easy to follow the structure of a block. I thought for a while about
how I might implement something like this in C++11, and eventually accomplished
this in two different ways.
My favorite of the two implementations. Allows for a synchronized block on any pointer using the syntax:
synchronized(&obj) {
// ... critical section
}
Note that this could have been done using an object rather than a pointer with
something more resembling Java synchronized(obj)
but it seemed to be more in
the spirit of C++ to use the address.
This is accomplished with a #define
, and a for
loop that initializes an
object to lock and unlock a std::mutex
on for
entry/exit. The mutexes
themselves are stored in a global std::unordered_map
of
void *
to mutex
.
An alternative use is a tablesynchronized
block. This requires the
programmer to create a SyncTable
and pass that as the first argument to
tablesynchronized
SyncTable table;
// ... later on
tablesynchronized(table, &obj) {
// ... critical section
}
The idea here is a group of threads can use a local SyncTable
rather than
the global one to avoid contention on the global table.
A non-intrusive use of multiple inheritance. A "lockable" version of any class can be instantiated with
Lockable<Object> lockable_object(/* Object ctor args */);
Any lockable object can be used within a synchronized
block. This approach
avoids the SyncTable
s entirely
synchronized(lockable_object){
// ... critical section
}
Can then be used without the overhead of a lookup for the mutex. Additionally,
no &
is used for lockable objects.