19 comments

[ 10.2 ms ] story [ 38.8 ms ] thread
No structs, just an array that accomplishes the same thing, without field names or other niceties. Enjoy the pleasure of not using a struct when you inevitably add/reduce/reorder fields later.
Good point! Added `vec_ptr[vec]` and `vec_len[vec]` helpers to mitigate this somewhat.
capacity isn't stored at all. Instead, it's computed on demand when the length of the vec is either zero or a power of two.

Brilliant insight. This is the first time I've seen this observation in over 3 decades of working with C.

Thank you! I implemented this pattern initially about a year ago, but decided to write about it now. Since you liked it, it wasn't for nothing. :)
That's pretty clever code. Too clever for my tastes.
(comment deleted)
This is just silly. You can't even reserve capacity because you only store size and capacity is implicitly the next power of 2 >= size.
(comment deleted)
Strictly speaking, the capacity is still stored internally to the allocation (it needs to be, in order to implement realloc)
Enjoy the annoying-to-debug errors when someone inevitably mixes arr[0] with arr[1] and tramples the heap (this could be mitigated by accessing the fields with macros), or writes arr[3] because they forgot this is not a regular array.
Yeah, I've added macros `vec_len` and `vec_ptr` (or, enums, actually), which help somewhat.

`arr[3]` should be flagged by the compiler it is known to the compiler that you're operating on an array.

You can pass `arr` as `&arr` to functions, then compiler will know the length of the array since the type would be `T ()[2]`.

And you can then use it like this:

  void f(int *(*ints)[2]) {
      for (size_t i = 0; i < (size_t)vec_len[*ints]; i++) printf("f: %d\n", vec_ptr[*ints][i]);
  }
Curiously, this is a rare case where the "inverted" `a[b]` requires less typing compared to `(b)[a]`.

A compiler will not be able to flag `vec_len[ints]` though, which is unfortunate.

> First of all, structs aren't used so you don't have to invent names for them (e.g. there is no IntVec)

But since it’s storing a void pointer any way, they wouldn’t need separate names right? You could use one struct everywhere regardless of the type of the items

Which IMO is a better idea than using an array here because the fields can be properly named and typed to prevent accidental misuse

Why would you want to avoid using a struct? Add a macro that declares the appropriate struct and get at least a tiny bit of type checking.

With some clever use of _Generic you could even build specialised functions for that type and get pretty good type checking

The most idiomatic and elegant 'dynamic array in C' solution is stb_ds.h, it's as simple as that :)

The 'public handle' is a pointer to the array elements so it has the same semantics as a regular C array, the meta-data (capacity and length) are stored directly in front of the array items. Growing the array has the same behaviour as realloc (e.g. you may get a new pointer back).

Thanks for mentioning it, I've heard about stb, but this time around I actually looked at the code. Their approach is very nice.
The C standard doesn’t guarantee that arbitrary integer values converted to a pointer and back result in the same integer values again. It only guarantees the other direction, that a valid pointer to void, when converted to uintptr_t and back again, will result in a pointer that compares equal to the original. The conversion from uintptr_t to pointer may for example clear or truncate some of the bits of the integer value, or normalize it in some other way.
You're right, I added some more clarity in the README.

Interestingly, I think my approach will work fine on CHERI, since the pointer is never dereferenced, but I didn't test this. But yeah, there are some architectures where it would fail.

Clever!

I don't like the use of uintptr_t, though. Why not storing the array begin in v[0] and a pointer to the first free item in v[1]? You avoid casts by defining len as ptrdiff_t and calculating it as v[1]-v[0].