quarticcat/enum-ptr

Ergonomic tagged pointer

Hi, it's QuarticCat

Projects

• detypify: Typst symbol classifier
• paguroidea: A reimplementation of the Flap parser in Rust

Notable Contributions

• difftastic (#393, #395, #401, blog)
  Wrote a bunch of optimizations, boosting it by 4x. (partially merged)
• lalrpop (#814)
  Rewrote the lexer, making the whole parsing process to be 2x~30x faster.
• exllama (#275)
  Optimized the hottest CUDA kernel, gaining 20% speedup on 2070S and 10% speedup on 3090. (note it's already well optimized)

Interests

• Algorithms and Data Structures
• High Performance Computing
• Functional Programming
• ... (learning more)

Languages

• Mostly: C++, Rust, Zsh
• Sometimes: Python, Javascript, Haskell

Find Me On

Blog (en/zh) • Stack Overflow • Codewars • Telegram Channel (zh)

enum-ptr's Issues

Enum Ptr 0.2.0 Plan

• Remove unit variant support and named variant support
• Support generic type params
• Change Aligned::Pointee to Aligned::ALIGNMENT
• Support Rc and Arc
• Rename feature std to alloc
• Rename map_ref_mut to map_mut
• Add abilities to use high bits
• Add more methods to smooth user experience
• Deal with the new discriminant value grammar

Return references from `map_ref` ?

Hi, thank you for making this crate.
I'd find it very useful to return refernces from map_ref, like this:

struct Bar {
    inner: u32,
}

#[derive(EnumPtr)]
#[repr(C, usize)]
enum Foo {
    A(Box<Bar>),
    B(Box<u64>),
}

let boxed = Foo::A(Box::new(Bar { inner: 1u32 }));
let compact_box: Compact<Foo> = boxed.into();

let reference = compact_box.map_ref(|foo| match foo {
    Foo::A(a) => &a.inner,
    Foo::B(_) => todo!(),
});

assert_eq!(*reference, 1u32);

I don't know if this is possible to achieve without UB, but it would be nice for sure.

Soundness issue of `map_ref_mut`

Consider foo.map_ref_mut(|f| *f = new_f), it will drop the old f. If the old f contains a Box, it will cause a double-free issue. Even f is trivially droppable, the modification won't be applied to foo, the compact f. We need to find a way to prevent users from modifying f itself or mark this function as unsafe or invent a new API to replace it.