Use perfect forwarding for StringObject::create

[herwinw]

This removes all the specialized implementations of StringObject::create with a generic one that just forwards the arguments to the actual private constructors. We can do the same thing with all other classes that have these create factory methods.

Small caveat: perfect forwarding is not perfect. The callers that use a stack allocated array don't work, these have been converted into calls using TM::String::create_and_take_ownership which removes a memcpy too.

[seven1m]

Well ain't that fancy!

Does the template programming slow down compilation at all? I do worry that using template programming seems to be slower (at compile time), at least with gcc.

[rolandpj1968]

Well ain't that fancy!

Does the template programming slow down compilation at all? I do worry that using template programming seems to be slower (at compile time), at least with gcc.

This is an interesting question when running at -O3. Legend has it that -O2 is a better balance of compilation time vs perf.

But no inlining? Dunno what clang and gcc actually do nowadays.

I'm actually surprised that member fields can be forward-referenced in header files. My C++ foo is a bit rusty.

In any case we could definitely factor some of the instruction->c++ generation in order to cut parsing at least.