asdl/format.py has heap-allocated BufWriter, which leak

[andychu]

Because the internal data_ pointer is malloc'd and we never call BufWriter::reset() which calls free()

[andychu]

This is exposed by mycpp/examples/parse.py

[andychu]

This one is good because it has a test with a simple verification

[andychu]

• the whole idea of having a data_ pointer that's not GC managed on a GC managed object can be problematic
  • data_ is just a heap-allocated Str

- Sweep -> free() 
- AllocStr(2 * old_size_)
- memcpy() 

[andychu]

@CoffeeTableEspresso

got rid of some leaks by explicitly calling reset(), also needed to change mycpp/mylib.pyi

Is this right architecturally?

[andychu]

@CoffeeTableEspresso I think I figured out a principled solution in the shower (of course)

It's not valid under Cheney though (because you can't really realloc with a bump allocator, although I guess you can simulate it ?)

• class BufWriter should hold a new type class Buf, which is like Str but mutable
  • the data_ or buf_ pointer should have a field mask ! Forgot this part
• So Buf is a full fledged GC object
  • A sequence like this should be valid: malloc(), realloc(), realloc(), realloc(), free()
  • one issue: we call calloc() to ensure it's zero'd.

---

Also to fix the quadratic problem it should be

realloc(10), realloc(20), realloc(40)  realloc(80) -- doubling

not

realloc(10), realloc(20), realloc(30)  realloc(40) realloc(50) ...

Basic knowledge that I've never really needed to use :)

[andychu]

Or actually an easier option is to make BufWriter itself like Str -- it uses the "flexible array" pattern like char buf_[1] ? And then placement new?

Although this could complicate generated code ... It couldn't use Alloc<BufWriter>, it would have to use a custom NewBufWriter

[andychu]

Also we zero'd for Cheney and string NUL termination of C strings

But this is slightly harder with realloc(), so maybe we can get away with just doing NUL termination and not zeroing the whole buffer

https://stackoverflow.com/questions/2141277/how-to-zero-out-new-memory-after-realloc

[andychu]

@CoffeeTableEspresso fixed this!