cehteh

'Segments' should become a newtype w/ trait too. Rationale: MemConfig determines the segment_size() arithmetically, but Vec / slice based storage backends could query that without calculation, only my planned thin...

another note mostly for myself, but opinions would be welcome: while the 'SegmentCache' try was working it turned out to be way to intrusive to be useful and 'Linear' outperformed...

instead dedicated feature this could be just enabled with: `#[cfg(debug_assertions)]` I see its arguable to add more complexity, but personally I would be in favor to add it, as it...

On 2023-07-15 15:42, Jacob Ryan McCollum wrote: >> * [ ] ThinVec proved bad, could/should be removed. >Agreed. I can do that, but will take some time. Being busy here...

On 2023-06-10 12:31, Jacob Ryan McCollum wrote: >@mccolljr commented on this pull request. > >Looks quite nice so far. Just a couple thoughts based on a first pass > >>...

I updated the benchmarks (not committed yet) and want to share some observations: 1. https://public.pipapo.org/segvec_bench/criterion/push%20values/report/index.html These `push()` 100k values. * In most cases we are less than 2x slower than...

This might closed now, the code exists as proof-of-concept. It could be reasonable to revive this idea when (cacheable) fast indexing on Proportional or Exponential is required. Still chosing Linear...

Note: I should've made this WIP/RFC > I believe the API is sound, but the implementation isn't. The `ManuallyDrop` needs to be created directly around the unerased pointer to avoid...

Fixed the ManuallyDrop/Guard issues. Still left the functions in the trait, I'd like to what you prefer, either making `ErasedPtr` a newtype as I suggested above or make `with()/with_mut()` toplevel...

To the above I'd like to add the following Brainstorming (eventually, if approved, I can implement this and make a PR) 1. Implement ClosingFD as above 2. Let 'ipc::channel()' return...