Skip to content

Instantly share code, notes, and snippets.

View thomcc's full-sized avatar

Thom Chiovoloni thomcc

View GitHub Profile
thomcc /
Created November 30, 2021 17:57
quick and dirty impl of lamports bakery algorithm in rust as an example. untested.
//! Implementation of [Lamport's bakery algorithm][bakery]. This is somewhat
//! interesting, because it's a mutex which can be implemented even if the
//! target only has atomic load/store, and no CAS (in principal it can be even
//! more general than this).
//! [bakery]:
//! Major caveat: This is not tested, and this algo is no longer appropriate
//! for modern code. Some variations of it can be useful in a thread pool, but
//! it's mostly useful to understand the concepts.
thomcc /
Created October 29, 2021 17:33
The old "magic git submodule wrangling" script from imgui-rs.
use anyhow::Result;
use std::{
path::{Path, PathBuf},
sync::atomic::{AtomicBool, Ordering::Relaxed},
/// "Automagic" submodule wrangling (sync, clean, fix, etc)
/// Ported from rustc's `` code, which runs something like this
/// automatically (even for `./ --help`)
thomcc /
Created August 3, 2021 21:55
alternative (more efficient, but still sound) way to implement `zeroize`'s memory wiping
//! Sketch of a alternative way to implement the backend of `zeroize` (the part
//! that does memory zeroing).
//! It should be:
//! - More efficient, possibly much more (for byte arrays), as it can zero in
//! the most efficient way available, rather than e.g. being forced to
//! perform element-at-a-time writes (where "element" often is "byte").
//! - Still guaranteed even in the face of aggressive compiler optimizations
//! as the semantics of volatile loads forbid the compiler from being able
pub type Argb8888 = u32;
/// Average of 2 pixels, rounding up on tie.
/// It's fine if they're actually BGRA8888 or any other format,
/// this treats every channel the same.
pub fn pixel_average(c0: Argb8888, c1: Argb8888) -> Argb8888 {
(c0 | c1).wrapping_sub(((c0 ^ c1) >> 1) & 0x7f7f7f7f)
// version that rounds down:
// (c0 & c1).wrapping_add(((c0 ^ c1) >> 1) & 0x7f7f7f7f)
thomcc /
Created July 9, 2021 00:38
Extracting the message from a `std::thread::Result::Err` (e.g. a `Box<dyn Any + Send>`).
use std::any::Any;
pub fn extract_message(err: &(dyn Any + Send)) -> String {
fn extract(err: &(dyn Any + Send), max_tries: usize) -> String {
if let Some(&s) = err.downcast_ref::<&'static str>() {
return s.into();
if let Some(s) = err.downcast_ref::<String>() {
return s.into();
//! initial version courtesy of @danielhenrymantilla.
//! extended to support fns with args/return types,
//! visibility, #[attributes], unsafe, const, async,
//! extern "abi" ...
//! left as an exercise for later (cuz it sux):
//! - generics?
//! - where clauses?
//! - probably other shit im missing
/// Something like a `flat_map` over a fallible iterator.
/// Note that the `map` itself is not fallible.
/// See test at end of file for usage.
// TODO this should probably take `impl IntoIterator<IntoIter = I>` but the
// generics here are already so fucking complicated, and i'm worried that
// will cause inference errors
/// Wrapper around an `impl std::error::Error` that makes it "mostly" clonable.
pub struct ClonableError<E> {
error: Option<E>,
// tbh these are all anybody cares about for Errors anyway.
display: Box<str>,
debug: Box<str>,
impl<E> Clone for ClonableError<E> {
thomcc /
Created April 2, 2021 10:08
bash function to resolve a crate version in a terrible way
# helper fn that returns version number given $crate or $crate/$semver.
# both must be url-encoded but the / must be a / if provided.
# doesn't handle cases where the version/crate/whatever doesn't exist.
resolve_version() {
# Perform `HEAD` request to `$1` and dump the redirected URL.
# We expect this to go to `$crate/$version`
# or `$crate/$version/$crate/` depending on if the crate has
# docs or idrk. Maybe a couple other urls are possible, it
# took a bit of testing for the `sed` script to handle all cases I found.
// type Int = i128;
type Int = i64;
type Rat = (Int, Int);
/// Quick and dirty (no overflow handling) continued fraction-based
/// rational approximation.
/// Returns `(p/q, was_exact)` where p/q is an approximation of `n/d`,
/// where neither `p` nor `q` are above `limit`. `was_exact` is true if