On Thursday, 27 June 2019 at 17:00:01 UTC, Rémy Mouëza wrote:

I though support for C++ allocation had improved. In a recent release, there was the addition of core.stdcpp.new, but I didn't try it out:
- http://dpldocs.info/experimental-docs/core.stdcpp.new_.html
- https://en.cppreference.com/w/cpp/memory/new/operator_new

It seems at this moment these operators only supported on Windows.

Are there still pitfalls to be wary of?

Not possible:

- new/delete (not entirely impossible, but requires to dig in compiler/library internals, generally fragile and not portable, not even mention that deciphering STL code is just meh...), though many libraries provides their own allocator/deallocator functions and some even provides pluggable hooks to use.
- Member functions pointers
- Multiple inheritance (tbh this is considered a bad practice anyway and chances to encounter it relatively small)
- lambdas?
- Exceptions? (except maybe on Windows, because of SEH)
- Template instantiations code gen(this basically requires to implement entire C++ compiler), what this means is that if the concrete template instance isn't used in a library you use you have to make dummy C++ function that forces compiler to emit the code for it.

Possible, but annoying:

- Functions with ref parameters

- Functions with const pointers to mutable data parameters
  ex: float calcStuff(float * const arr, size_t len);
It doesn't make sense in D, and I doubt there is much sense in C++ as well (except maybe to convey the meaning that this function won't try to free data). So the workaround of course is to slap pragma mangle, which of course requires manually getting the mangled name for the function...

On Windows using MS compiler for the example above the mangled name will look like
  float calc(float * const arr); // ?calc@@YAMQEAM@Z
and for non const
  float calc(float * arr); // ?calc@@YAMPEAM@Z

so it is possible to do something like this
  pragma(mangle, calc.mangleof.replace("QEA","PEA"))
  float calc(float * arr);

- Some operator overloads also needs pragma mangle treatment.

- Incomplete/buggy mangling support, especially annoying with templates. Same treatment.

Maybe I've missed something else though, but anything not on the list usually works without surprises. And thanks to string namespaces it is now such a relief to use.

Reply via email to