auxoncorp / fixed-slice-vec Goto Github PK
View Code? Open in Web Editor NEWVec with dynamic length and runtime-determined capacity backed by a slice. no_std , no-allocator friendly.
License: Apache License 2.0
Vec with dynamic length and runtime-determined capacity backed by a slice. no_std , no-allocator friendly.
License: Apache License 2.0
I would say it would be incredibly useful to have such alignment function in cases where multiple vectors are to be allocated.
For instance, here, in our memflow project, we perform a series of slice splits to allocate the buffers, with intention of allocating that particular number of elements in each, rather than "just how many fit in the misaligned buf". This is particularly troublesome here, where the allocated buffer is expected to fit exactly that many elements. However, in some cases the split slice is misaligned, and the resulting capacity is zero (thus I had to add extra bytes in the split).
A with_capacity_align_from_uninit_bytes
function (a really long name, would probably need something simpler) would try to align and split the slice to house just that particular amount of elements, while keeping the end to be freely used by other vecs.
If you think this is an useful feature, I could go ahead, add it in, and create a PR.
I'm curious what you found that prevents the arrayvec crate itself from being modified to support both array and slices? I suppose the compiler might fail to propagate the constant length somewhere, causing a performance regression relative to arrayvec?
I just encountered use-case where your crate can be suitable but noticed few things that should be improved.
I suggest to add additional bound T: Drop
here because it would be noop anyway after optimization.
With new bound, rustc wouldn't generate drop calls when FixedSliceVec with Copy types goes out of scope so it would have less compilation errors.
Also, it would make programs like that compile (currently you need to manually drop v before second loop):
let mut arr: [MaybeUninit<u32>; 50] = MaybeUninit::uninit_array::<50>();
let mut v: FixedSliceVec<u32> = (&mut arr[..]).into();
for i in 0..10{
v.push(i);
}
for v in arr[..10].iter(){}
FixedSliceVec::clear
and FixedSliceVec::truncate
can double dropCode here
Line 291 in df78d16
If vec contains few items inside and item in the middle panics during dropping, you can call clear
again and make double-free for first items in vec.
I suggest to move length zeroing before unsafe block so you would just leak data in case of panic.
Another option is to call pop
until our len is 0 but it can be less performant and would make API not clear (e.g. we can still have items after panic).
Same can happen in truncate function.
FixedSliceVec::pop
can be improvedIt would be easier for optimizer to understand code like that:
self.len -= 1;
/*return*/ Some(unsafe { self.storage[self.len].as_mut_ptr().read() } )
It uses MaybeUninit
utilities instead of slice trick.
It would be useful to be able convert vec to initialized part of inner slice if elements is Copy. Since elements is Copy they cannot implement Drop so we don't need to worry about leaking elements.
Use case:
/// Returns all items which was processed.
fn do_some_processing_for_items<'a>(items: &[u32], buffer_for_output: FixedSliceVec<'a>)->&'a [u32]{
for item in items.iter(){
if can_process(item){
process_item(item);
buffer_for_output.push(*item);
}
}
buffer_for_output.into_initialized_slice()
}
This crate needs dramatically more tests and a greater variety of tests.
miri
testing
- Create an array of 3
String
.- Initialize a
FixedSliceVec<'_, String>
with it.- Call
clear
, all 3String
are dropped.- At the end of the function, the array is dropped, dropping the
String
again.
As noted by @HeroicKatora , these constructors mean that when a user writes a type with padding bytes into the backing slice, and then the FixedSliceVec/single is released, those padding byte contents are visible in the original backing collection, which is undefined behavior.
At very least this deserves requiring users to opt-in to managing the appropriate safety concerns.
A declarative, efficient, and flexible JavaScript library for building user interfaces.
๐ Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.
TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
An Open Source Machine Learning Framework for Everyone
The Web framework for perfectionists with deadlines.
A PHP framework for web artisans
Bring data to life with SVG, Canvas and HTML. ๐๐๐
JavaScript (JS) is a lightweight interpreted programming language with first-class functions.
Some thing interesting about web. New door for the world.
A server is a program made to process requests and deliver data to clients.
Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.
Some thing interesting about visualization, use data art
Some thing interesting about game, make everyone happy.
We are working to build community through open source technology. NB: members must have two-factor auth.
Open source projects and samples from Microsoft.
Google โค๏ธ Open Source for everyone.
Alibaba Open Source for everyone
Data-Driven Documents codes.
China tencent open source team.