Crate of the Week

I self nominate pm100/dyncall

dyncall allows interpreters, script based languages to dynamically call arbitrary functions in shared libraries. The repo references sample implementations for basic, forth and lox (all forked from other peoples work). Here's a basic sample

10 def xfn msgbox("user32.dll|MessageBoxA|ptr,cstr,cstr,u32|i32|")
20 print "Showing a Yes/No dialog..."
30 let result = fn msgbox(0, "Do you like dyncall?", "dyncall BASIC demo", 4)
40 if result = 6 then print "You clicked Yes!"
50 if result = 7 then print "You clicked No!"
60 print "MessageBoxA returned "; result

the 'def xfn' is an extension I added to the basic to define external functions. This calls into dyncall, the later msgbox call uses dyncall to push the arguments and perform the function call.

lox playing sounds

// Sound demo using mciSendStringA via dyncall
// mciSendStringA(cmd, retbuf, retbuflen, hwnd) -> u32

var mci = exfun("winmm.dll|mciSendStringA|cstr,ptr,u32,ptr|u32|coerce");

print "Playing 4 sounds...";

mci("open C:/Windows/Media/tada.wav type waveaudio alias snd", nil, 0, nil);
mci("play snd wait", nil, 0, nil);
mci("close snd", nil, 0, nil);

same demo in forth

\ Sound demo using mciSendStringA via dyncall
\ mciSendStringA(cmd: cstr, retbuf: ptr, retbuflen: u32, hwnd: ptr) -> u32

extern: mci "winmm.dll|mciSendStringA|cstr,ptr,u32,ptr|u32|coerce"

\ Helper: send an MCI command and drop the return value
: mci-cmd ( str -- )
0 0 0 mci drop ;

." Playing 4 sounds..." cr

zstring" open C:/Windows/Media/tada.wav type waveaudio alias snd" mci-cmd
zstring" play snd wait" mci-cmd
zstring" close snd" mci-cmd

works on windows , mac, linux. All example so far are windows, heres a linux example

10 REM Standard stream example - Linux/macOS version
20 REM Use fopen("/dev/stderr", "w") to get a FILE* for a standard stream.
30 REM /dev/stdin, /dev/stdout, /dev/stderr are always available on Linux and macOS.
40 REM On Linux use libc.so.6; on macOS use libSystem.B.dylib.
50 PRINT "Standard stream example (Linux/macOS)"
60 DEF XFN c_fopen("libc.so.6|fopen|cstr,cstr|ptr|")
70 DEF XFN c_fputs("libc.so.6|fputs|cstr,ptr|i32|")
80 DEF XFN c_fflush("libc.so.6|fflush|ptr|i32|")
90 DEF XFN c_fclose("libc.so.6|fclose|ptr|i32|")
100 REM Open stderr and stdout
110 LET fp_stderr = FN c_fopen("/dev/stderr", "w")
120 LET fp_stdout = FN c_fopen("/dev/stdout", "w")
130 REM fputs to stderr writes directly to the console
140 LET r = FN c_fputs("Hello from BASIC via fputs (stderr)", fp_stderr)
150 PRINT "fputs returned: "; r
160 LET r = FN c_fputs("Result of 6 * 7 = 42", fp_stderr)
170 PRINT "fputs returned: "; r
180 REM fflush stdout to ensure buffered output is written
190 LET r = FN c_fflush(fp_stdout)
200 PRINT "fflush stdout returned: "; r
210 LET r = FN c_fclose(fp_stderr)
220 LET r = FN c_fclose(fp_stdout)
230 PRINT "Done."
240 END

I'd like to self-nominate seqpacker, a bin-packing library for packing variable-length sequences into fixed-capacity bins β€” built for LLM training, but usable for any bin-packing problem.

  • 11 algorithms (NF, FF, BF, WF, FFD, BFD, FFS, MFFD, OBFD, parallel OBFD, Harmonic-K) behind a uniform API
  • ~5 ms for 10,000 sequences β€” ~1.2x faster than the closest C++ alternative, 400–1900x faster than pure Python
  • Pure Rust core with #![forbid(unsafe_code)], Python bindings via PyO3
  • Prebuilt wheels for Linux, macOS, and Windows β€” no build dependencies for Python users

Links: GitHub | crates.io | docs.rs | blog post

My suggestion: ailed-soulsteal β€” a fast CLI tool that tokenizes PGN chess files into compact binary training data for ML pipelines. Processes 60K+ games/sec (1M games in ~15 seconds) using streaming parsing. MIT licensed.

Blog post: Processing 1M Chess Games in 15 Seconds with Rust - DEV Community
GitHub: GitHub - Ailed-AI/ailed-soulsteal: Fast game data extractor β€” drains PGN into tokenized training data for AILED-Soma. Named after Alucard's Soul Steal from Castlevania: Symphony of the Night. Extracts, filters, and tokenizes turn-based game records at C speed. Β· GitHub

I'd like to self nominate skilo. A crate for Agentic AI skills management, linting, and formatting

The nomination text

If I may nominate a binary crate, I suggest koharu - Tauri app with end-to-end pipeline for manga translation.
Its pipeline is four things, specifically: locating text which has to be translated in an image, OCR-ing it, suggesting a translation by invoking a local LLM, and then rendering it over the original image.

What's notable is that all it is local, using llama.cpp bindings; also a body of work was used for smoothing out corners like vertical text rendering, system keyring integration, etc.

GPL3 licensed, which is somewhat uncommon.

I revoke and can no longer endorse it because it started collecting telemetry, potentially including user IPs (on April 8th).

I’d like to self-nominate burn-speech-training β€” an end-to-end speech model training pipeline for the Burn ML framework. Includes MFCC extraction, CTC loss training, LibriSpeech loading, and evaluation against human labels. First speech training example I know of in Burn.

I'd like to self nominate acdc-parser, an AsciiDoc parser aiming for supporting the full AsciiDoc language.

2 Likes

I would like to self-nominate aimdb-core, a type-safe and platform-agnostic data pipeline where the Rust type system is the schema and trait implementations define its behavior. Enforce unified data contracts from silicon to cloud β€” same codebase on no_std MCUs, edge gateways and Kubernetes.

8 Likes

I'll self-nominate my latest CLI tool "qb", an interactive terminal user interface for Kubernetes (I call it EMACS for Kubernetes).
https://crates.io/crates/qb

It has all features that I need on a daily work with Kubernetes - same ones that you know from UI tools incl. logs, (persistent) port-forwarding, multiple profiles, edit capability, cluster overview & resource map w/ utilization, batch job runs etc. etc.
I found the "related resources" particularly helpful - this let's you navigate from a pod to mounted secret to other pod to [more] ...
Most importantly it also features a context-aware help menu to learn all interactions as they've grown quite drastically.

Screenshot shows an empty kubeadm with metrics server running locally as example.

1 Like

I’d like to nominate my crate, speech-prep (crates.io: Rust Package Registry), for Crate of the Week.

It’s a focused Rust crate for speech audio preprocessing: format detection, WAV decoding, sample conversion, preprocessing, chunking, and VAD. I extracted it from a private codebase and cleaned it up into a standalone OSS crate with a deliberately narrow scope.

Repo: GitHub - dnvt/speech-prep: Speech-focused audio preprocessing β€” VAD, format detection, decoding, noise reduction, chunking Β· GitHub

Join the official launch of rust-key-paths library Today.

crate:
https://crates.io/crates/rust-key-paths
https://crates.io/crates/key-paths-derive

(post deleted by author)

I'd also like to nominate my own crate (x360-w-raw-gadget): crates.io: Rust Package Registry

This is a crate for emulating an Xbox 360 wireless receiver on Linux via the raw-gadget kernel interface. It supports up to 4 controller slots, exposes a C FFI layer for Python/ctypes use, and has been tested on a Raspberry Pi Zero 2W. Still a work in progress, but it fills a gap that standard USB gadget interfaces like FunctionFS can't. Xbox controllers are widely supported on all kinds of drivers, so this allow for some nice glue code with any usb otg enabled sbc

(accidental reply-post above, new to this platform)

1 Like

(post deleted by author)

I'd like to self-nominate Myth Engine, a high-performance, cross-platform rendering engine built from the ground up in Rust.

It features a declarative render graph, PBR pipeline with HDR, glTF 2.0 support with animation, and Python bindings for integration with AI and scripting workflows.

Demo: Myth Engine β€” Showcase
Repo: GitHub - panxinmiao/myth: A High-Performance Rendering Engine for Rust Β· GitHub

1 Like

Self-nominating bincheck (GitHub - kazu11max17/bincheck: Binary hardening checker for ELF and PE β€” RELRO, NX, PIE, CFG, and more Β· GitHub).

A CLI tool for checking binary hardening properties on ELF, PE, and Mach-O files.
Inspects RELRO, Stack Canary, NX/DEP, PIE/ASLR, CFG, and more.
Outputs results as table, JSON, or SARIF for CI integration.

Useful for embedded/firmware security audits where you want to verify
compiler hardening flags are actually applied to your binaries.

flodl - A deep learning framework built on libtorch. Up to 31% faster than PyTorch, heterogeneous multi-GPU DDP, and a Graph computation builder.

1 Like

The Embassy Executor - Crate of the Year :slight_smile:

I’d like to share my project, Celerity a pure Rust implementation of ZMTP 3.1 built around a sans-IO core. It provides Tokio-based TCP/IPC transport, PUB/SUB and REQ/REP helpers, plus optional CURVE-RS encryption for secure remote links. Designed for developers who want a lightweight Rust-native messaging stack without depending on C libraries.

heres the github link:GitHub - ShemShadrack/celerity: Pure Rust sans-IO ZMTP 3.1 engine with Tokio TCP/IPC transport and CURVE-RS Β· GitHub