> > * If your compiler supports C++17, this should work without any external > > dependencies(apart from APR). Other compilers will require boost. > > Great and obviously a lot of work! Though, I've ran into multiple > problems with my old C++-Builder 10.2 UPD 3 with 32 Bit BCC-compiler. > "boost::atomic" doesn't seem to be available and you are using > language constructs not supported yet, like the following: > > https://stackoverflow.com/a/2795024/2055163 >
That was one area that I wasn't sure of, if that would work with your system. The atomics are also not strictly needed, but they provide a convenient lock-free way of setting variables without locking the mutex. I ran into a number of problems with recursive mutex locking, so atomics sidesteps that issue. It's also likely slightly faster, since you don't need to lock and unlock the mutex, although with a modern system it's probably not noticeable. > It seems worth it discussing (again) if such old compilers should be > supported at all anymore or not. I'm not sure if there was a concrete > result in the end sayong only C++11 and newer or alike. > I feel that requiring at least C++11 is perfectly reasonable at this point. The latest release is a good choice for legacy systems, and if needed we can branch from the last release(or from the current master version) to give updates for those platforms. New code is also much easier to implement with newer C++ features, and has the potential for better optimization using move semantics for example. There are a few features that BCC32X doesn't support from C++11, but those should be easy enough to work around(thread_local being an example). Some other compilers took a while to implement the C++11 features, although those should be pretty rare at this point(e.g. MSVC before 2015). Oddly enough, the newer C++ versions seem to have better support in older compilers(e.g. they were implemented by compiler vendors faster than C++11). Requiring C++11 also simplifies some code - as you saw, I left in some std::shared_ptrs around instead of log4cxx::pointer, but the code that detects for C++ wouldn't be needed if we can assume that it already exists(some detection code for C++14/17 features would still be required). -Robert Middleton
