Reiner Pope

Results 12 issues of Reiner Pope

The documentation on `double` states that it loses accuracy on some inputs, but using `rational` will fix this. I understood this to mean that if one simply used `rational ::...

question

For some libraries I'm writing, I'd like to provide multiple versions of my functions with support for static dispatch, but I don't want to take a dependency on `syn`, and...

For ``` #[repr(transparent)] struct Id(NonZeroU32); ``` this type is indeed zeroable in Option, but implementing it requires the user to write the word `unsafe`. Would be helpful for `bytemuck` to...

`jax.make_array_from_callback` does a sequential loop over devices attached to this host: https://github.com/google/jax/blob/ef40b85c8b2686f64bc9ca67de267a6b1a7935bb/jax/_src/array.py#L693. When fetching from remote storage with high latency, this sequential loop can become latency-limited rather than throughput-limited on...

enhancement

## Problem In the `ToLexical` APIs, they take the destination buffer as type `&mut [u8]`. According to Rust's soundness rules, this requires the buffer to be initialized before calling `to_lexical`...

enhancement
high priority

Instead of varlen encoding, which requires up to 8 conditional branches in a loop, we use a simpler encoding which is just 4 instructions without branching. New encoding: * length...

This SAFETY comment (https://github.com/Robbepop/string-interner/blob/95574e27c0cda19113ab27dc1e8a5f9f9fa5ab8a/src/backend/buffer.rs#L102) seems to me to be reasoned incorrectly. It seems to be true for values of `index` that were returned by this `BufferBackend`, but when `index` is...

I love jaxtyping! Can I have more of it please? Specifically, I'd like to make assertions about the sharding of my `jax.Array` objects. Given an array `Float[Array, "batch seqlen channel"]`...

feature

The `assert isinstance(...)` pattern prints a mostly useless message, just "AssertionError" without explanation. Would it be possible to expose an `assertIsInstance(x, ty)` API that prints expected versus actual, like we...

feature

If we were willing to break backwards compatibility, it would be possible to support deriving `Pod` for arbitrary generic types. The approach: * extend `trait Pod` to include `const POST_MONOMORPHISM_CHECKS_PASS:...