"Fearless Concurrency" predates "async" in rust nomenclature, and points to specific compile-time guarantees around things like thread safety and races (Send, Sync, ...).
It was around when we used event loops and the mio crate (and rayon for parallelism), with no async or await in sight (because they didn't exist yet).
Some people actually still prefer to do "concurrency" this way, especially if they wish to stay close to the abstraction level from C land. io_uring also came later which introduced a new OS low-level interface and paradigm for doing "concurrency".
"Fearless Concurrency" has nothing to do with how easy you can write async code, especially if you're someone who struggles with the language basics like ownership.
It's not mostly compatible, not even on the surface level, with any version of C post C89. And most of the ever-growing crap in the language came after the early years anyway, with constructs that are C++exclusive.
Not cargo per se, but even the tutorial for a cli-tool is like “setup clap, which has 20 dependencies and a kitchen sink”. The whole (cargo-centric) ecosystem is much like Node, with the same problems.
cargo new with-clap
cd with-clap
cargo add clap --no-default-features
The only correct way to package such software is to vendor dependencies (packaged together or separately). And you can trivially change the sonames of vendored deps in your build scripts so that there are no conflicts whatsoever (I dual-package some stuff against an upstream and a fork and do just that). So dynamic vs. static is not the crux of the issue. The primary concerns are that distributors hate vendoring, irrespective of whether the vendored libs are linked in statically or dynamically. Distributors also hate potentially diverging forks maintained by random downstreams, which is what "patched dependencies" effectively are.
There is always room for some leeway of course, but that would depend on how relevant your software is, and/or whether a maintainer would want to take that burden on.
And finally, sometimes, such dependencies may provide added value that trumps all these concerns. So judging these things is always situational.
Then you could be forced to vendor everything. And if it's open-source and relevant for distros to pickup, then you will need to find out if distros would be willing to take your library with its vendored libs (or package them separately just for your library)...etc.
And you may need to figure out if there are bus factor concerns with your direct dependency, since such libraries are not necessarily maintenance free, even from a mere compiling/building stand point (what if a patched indirect dependency no longer builds with new compilers...etc).
This would depend on the language/ecosystem. It's worse for C and C++ than for example Rust because of packaging policies and ease of distributability.
Off-topic: The switching between light background for text and dark background for code is stupid and annoying. Either use dark for both, or light for both (so I can apply my custom inverse shader without hurting my eyes or breaking the reading flow). This back and forth switching is almost as stupid as using a background color with middle lightness, which I've also seen.
This has been irking me for a while, so I decided to randomly whine about it now.
As for the blog content. It's a weird collection of clippy bugs which covers basically the first half of the blog, then a couple of miri bugs and limitations. Then it finally gets to actual rust compiler stuff which just involves cfg_select (a nice to have) and c_variadic support (ffi use-case). So, IMHO, that title is a bit over the top, and it almost does a disservice to a lot of actual good and interesting work that is being done.
Four different hex representations for the same 3 bytes is extremely wasteful. You don't need to store more than a [u8; 3]. You can create a "mapped" struct with those representations if caching them is somehow useful for the use-case. And since we're micro-optimizing memory usage, Box<str> instead of String (for the two fields with actual string data) would be more efficient since those values are not meant to be mutable. You can then serialize to one of the many efficient binary formats around (e.g. borsh). This is not JavaScript, so you can oftentimes not be bound to JSON.
Your parsing code looks both unrobust (potentially fragile hard-coded assumptions) and unnecessarily complex. Something much simpler can probably be done with some trivial regex usage.
--no-default-features --features=foo,bar is fine. The harmful part is replacing established crates with smaller ones solely because of ze size.
And the whole dance doesn't even do what many people think it does, as covered in some comments in that linked old thread.
Note that I made that jerk thread when min-dependency-ing was sort of trending. A trend that was often (not always) as stupid and counter-productive as the other related(-ish) trend min-binary-sizing.
Also note that the harmfulness of that trend went beyond bad ecosystem dynamics and dependants ending up with less quality/reliability. That semi-obsession also encouraged bad coding practices like pervasive use of dyn where it's not needed. Compromising idiomaticity, and removing zero-cost abstractions to win a faster compiling dependency prize!
I will stop here, because I really don't want to write that blog post.
Author couldn't help himself but do some "wisdomry" at the end!