Minimize memory allocations when assigning multiple values to string variables.

[Varriount]

Currently, whenever a string value is assigned to a variable, the string value is copied into a new piece of memory, regardless of the contents of the previous value. This can be optimized in certain cases - if assigning a value to an already initialized string variable, and the value being assigned is the same length or smaller than the already initialized string, than a simple memory copy can be done, rather than a dereference of the old string and creation of a new string.

Test case:

proc main =
  var a = "Hello"
  var b = "World"
  var c = "Hello World"

  echo(repr(a)) # Address 000000000037C050
  echo(repr(b)) # Address 000000000037C078
  echo(repr(c)) # Address 000000000037D050
  echo()
  a = b # Since the contents of B can fit into a, the contents
        # are copied and a new string is not created.
  echo(repr(a))     # Address remains unchanged.

  a = c # Since the contents of C *cannot* fit into a, a new string
        # is created.
  echo(repr(a))     # Address is changed.

main()

Though this means checking the length of the old string variable on each assignment, this overhead is probably much less than that of allocating memory for a new string and garbage collecting the old string.

[simonkrauter]

If I understand it correct, you can do this only when ref count is 1.

[Varriount]

@trustable-code Yes, this only applies to strings with a recount of 1. Thankfully, due to strings being value types, this is the case 95% of the time.

[Araq]

Interestingly, this could be a good idea for the upcoming strings based on destructors too.

[metagn]

We have moves now, in the example the addresses of b and c are both sustained when moving into a

[Araq]

Bites with COW-strings which hopefully will come to Nim, eventually.