[llvm-dev] RFC: Should SmallVectors be smaller?

[Dean Michael Berris via llvm-dev]

> On 22 Jun 2018, at 12:30, Bekket McClane <bekket.mcclane at gmail.com> wrote:
> 
> To Dean,
> 
> I think Duncan’s approach prohibit any usage of Small after the capacity grow over SmallCapacity.
> So when the capacity exceed SmallCapacity, one should:
> 1. Allocate memory on heap
> 2. Copy data from  Small to that chunk
> 3. Assign pointer of that chunk to Large
> 
> As long as you always access Large after growth, there would be no data lose.
> 

Ah, yes. That makes sense.

I’m curious to see, like Chris, what the benchmarks say about this alternative approach.

-- Dean