YAMA 0002 - Adding support for stack allocated structs

• Author(s): Bhathiya Perera
• Status : ✅

Problem

class Enemy:
    x: int
    y: int
    hit_points: int

def main() -> int:
    enemies: Array[Enemy] = arrnew("Enemy", 2)
    enemies[0] = Enemy()
    enemies[1] = Enemy()
    return 0

Given a structure like above, it makes no sense to allocate individual objects in heap all the time. We can avoid it by using Tuple however we lose the information about field names then. Keep in mind that the Tuple’s in Yaksha are mutable (Mutable Tuples are what we have at the moment, may or may not change in future).

Suggestions

Suggestion 1

@namedtuple
class Enemy:
    x: int
    y: int
    hit_points: int

• Pro:
  • Consistent syntax
  • Does not introduce a new keyword
• Con:
  • No way to tell that this get allocated in stack?

Suggestion 2

@dataclass
class Enemy:
    x: int
    y: int
    hit_points: int

• Pro:
  • Consistent syntax
  • Does not introduce a new keyword
  • Familiarity
• Con:
  • No way to tell that this get allocated in stack?
  • Looks like python but different might make it confusing

Suggestion 3

class Enemy:
    x: int
    y: int
    hit_points: int
Enemy()     # << allocate on stack
new Enemy() # << allocate on heap

• Pro:
  • Similar to how C++ works.
  • new - del pairing
• Con:
  • Need to rewrite everything to follow this convention
• Note⚠️⚠️:
  • Perhaps reconsider this approach at some point

Suggestion 4 ✅

@onstack
class Enemy:
    x: int
    y: int
    hit_points: int
# Rename @dotaccess to @onstack
# And use that for this purpose
Enemy() # This is disallowed. Just create a variable and you can use it.

• Pro:
  • Defines that this is different from a normal class
  • Clearly visible that this get allocated on stack
• Con:
  • Require changes to how @dotaccess works currently

Conclusion

I am thinking suggestion 4 is the best approach.

Update

Additionally Yaksha now supports struct keyword. That act as a syntax sugar for @onstack class.