deprecates NewFinalize which is incompatible with non-var destructor

[ringabout]

proc new*[T](a: var ref T, finalizer: proc (x: ref T) {.nimcall.}) {.
  magic: "NewFinalize", noSideEffect.}

The new function supports a finalizer with a ref T type, which is not compatible with the non-var destructor. Because in a finailizer, users are free to modify the parameter. You cannot wrap it without make it mutable in a non-var destructor, e.g.

proc delete(self: Foo) =
  self.data = 0

new(result, delete)

[Araq]

I don't understand this. Surely the finalizer can call destroy(x[])?

[ringabout]

type
  FooObj = object
  Foo = ref FooObj
     data: int

proc delete(self: Foo) =
  self.data = 0

new(result, delete)

the new finalizer creates a wrapper around the delete function

proc deleteHook(self: var FooObj) =
  delete(self)

Then deleteHook is bound to the FooObj. The non-var destructor requires deleteHook to be immutable, i.e., delete to be immutable.

[Araq]

That only means that new got more picky about the finalizer's requirements, not that it needs to be removed entirely.

[arnetheduck]

not that it needs to be removed entirely.

the fact that one call to NewFinalize changes the behavior for all instances of that type is a major wart - getting rid of this atrocity would do the language good.

[Araq]

the fact that one call to NewFinalize changes the behavior for all instances of that type is a major wart - getting rid of this atrocity would do the language good.

That is not an issue anymore with arc/orc anyway and it's now done consistently. Afaik.

[ringabout]

Succeeded by #24354