Hacker News Reader

Comment by The_rationalist | Hacker News Reader

It's fascinating how much the rust community dedicate work to reinvent the wheel. Don't get me wrong, it's not always a bad thing, but I believe there are still a load of interesting types of software that almost doesn't exist, and such efforts would have better ROI on those new kinds of software e.g an hypergraph database or an NLP library that allows for any inflected word to get another inflection (e.g I give in parameters input="be", kind=verb, time=present, person= 3rd plural) and it output "are".

View on HN

The ROI on improving widely used tools is quite large. And for most of them just being in a memory-safe language is already a large improvement.

Not everybody wants to work on hot new things, some want focus on having their tools to stable and working as designed. We need both kinds of work.

I agree and I disagree.

The ROI on improving widely used tools is quite large Yes indeed. You probably wanted to say: The ROI on reinventing widely used tools is quite large" There are some extremely used binaries that have a stagnant development and have not adapted to modern ergonomics. I beautiful example of that subset of widely used tools is cat. Cat is extremely used and yet, the rust rewrite bat is far superior, because among other things, it has color and code higlithing :0

For a rust rewrite to be successful you need the following conditions: You have ideas of improvments not present in the current tool. Your ideas are not on their roadmap. The current tool has not had too much of man/hours, which you can interpret heuristically as: the current tool has less than a few thousands of commits.

Here we're talking about systemd which has 42,600 commits. By the time rustysd will have as much features as current* systemd, the original developer will be dead.

As an end user I have almost zero benefits in this kind of rewrites, as I'm aware it's a far poorer software by design.

If it was 1) a new kind of software or 2) a reimplementation of a tool that they can realistically manage to outcompete or 3) the development that will in the very longterm replace an old complex tool that has stopped development, then I would have been interested.

Anyway if it's for fun/learning purpose, they do what they want, but utilitaristically there would be better choices.

> For a rust rewrite to be successful you need the following conditions: You have ideas of improvments not present in the current tool. Your ideas are not on their roadmap. The current tool has not had too much of man/hours, which you can interpret heuristically as: the current tool has less than a few thousands of commits.

I think you are missing the point completely. We are talking about rust, memory safe performant language, and systemd, core system in many Linux distributions, written in C++. I would love to be able to replace systemd (and kernel and core libraries too) with something that was safer at least in this aspect (memory safety).

Big kudos to Mozilla for identifying the cause of many security bugs and for finding a way to solve it while keeping performance intact.

Memory bugs are only one kind of bugs. Even if stability where our only metric, systemd had more testing and bugfixing than a rust alternative would ever have. Thus even if the rust one would be free of most memory bugs, the testing made on systemd would outweigh by far this advantage.

BTW by the time of the rewrite, c++ will have memory safety mechanisms similar to rust: https://news.ycombinator.com/item?id=18037566

While those C++ efforts are admirable, and good, they’re only setting out to do a subset of what Rust can do.

And Rust is only setting out to do a subset of what C++ can do. Did you have a point?

In particular, C++ has been engineered from the beginning to create expressive, powerful libraries, much moreso than Rust. The gap today is bigger than it was five years ago. This matters because every use of an idiomatic, powerful, mature library replaces code not using it that would have had bugs. Yes, Rust programs do have bugs.

Powerful is important in libraries because it translates to actually usable where you need it, without compromise. Any library you can't or don't use does you and your users no good.

Code that uses a good library will have none of the problematic, buggy constructs that Rust is supposed to protect you from.

There’s no need to be so aggressive.

By "aggressive", you mean explaining the point in detail?

Rust and C++ have different emphases. C++ has a huge head start, but also massive backward-compatibility anchors. It is certain that, over time, Rust will get better at expressing powerful libraries, but C++ is not sitting still.

It is very far from clear whether Rust will be important, ten years from now. It is far from clear that it is more beneficial to put resources into rewriting C things from scratch in Rust than to transition existing development from C to C++, and start replacing subsystems in-place.

What is clear is that the difference in benefit to starting any new project in Rust or in C++ is absolutely overwhelmed by the absolute benefit of starting in one of those, and not C.

> By "aggressive", you mean explaining the point in detail?

No, I mean the stuff dang has already commented on.

You also did not explain the point in detail, you responded to what I said with a whole bunch of off-topic things, and then insinuated that I did not have the point.

The parent said that C++ would have "memory safety mechanisms similar to Rust," and my point was that while they are similar in some ways, they are not as comprehensive. That's an important distinction!

All of this other stuff you're talking about has nothing to do with that.

What I see is special pleading. C++ inherits flaws from C, Rust has flaws from its late start and weaker emphasis on expressibility, both sets of flaws interfere with reliable use.

Please edit swipes like "Did you have a point?" out of your comments here.

https://news.ycombinator.com/newsguidelines.html

I am given no opportunity to edit it out.

It was an honest question, and was not answered.

The edit window expired, but if you want us to edit your comment for you, you'd be welcome to email the text you want to hn@ycombinator.com.

If you say that you mean it sincerely as a question and not as a swipe, I believe you. The problem is that lines like that are routinely used to take cheap shots at others, imply that they don't have a point, and so on. If that wasn't your intention, the burden is on you to disambiguate that by making clear that you're asking sincerely and making your question more specific.

> In particular, C++ has been engineered from the beginning to create expressive, powerful libraries, much moreso than Rust.

I really can't fault Rust developers for choosing to put aside the whole C++ ABI mess by refraining from developing their own "Stable ABI" and sticking to the lowest common denominator of C FFI for the time being. And perceived ABI issues are the only thing that might stop you from building "expressive, powerful libraries" in the language. (Const generics and specialization are often-requested features that do impact this area, but they're in the pipeline already, and coming soon.)

> "perceived ABI issues are the only thing that might stop you from building 'expressive, powerful libraries' in the language."

This statement makes no sense. What stops you from building expressive, powerful libraries in Rust is the absolute lack of the corresponding features in the language that are necessary to the task.

Oh, boy. Don't go there.

No amount of testing and bug-fixing solves badly created software. Systemd is begging for a rewrite.

Now, I don't know if this one will be complete enough to replace the original. But yours is not a promising argument.

Name one problem you personally have encountered deploying systemd in production.

The launching order of the internal daemons is not explicit on the services graph. For a long while I had a computer that mounted the disks on a non-predictable time and most attempts to fsck the root partition would fail.

For a while it would ignore the system DNS settings, so it didn't only require it to be configured twice, but the systemd settings had an habit of changing in format every so often.

Udev failed a bit at the time it was being included on the package.

Oh, there was the time when the simple service launcher didn't conform to the documentation so a lot of daemons could fail to launch in unpredictable ways.

But the first I started using it was at the Debian migration. So the earlier bugs were all solved already.

> systemd, core system in many Linux distributions, written in C++

s/C++/C

Yep, systemd is in plain old C....I wish it would be in C++....

Most of the security vulnerabilities reported (https://www.cvedetails.com/product/38088/Freedesktop-Systemd...) would not have existed in modern C++. Including all the heavy ones implying stack smash due to VLA buffers that do not exist in C++.

Many will hate me for that, but preferring old-C over C++ for a system software after 2011 is a non-sense. Even the maintainers of GCC understood that.

Indeed. It is other programs that are written in C++. (-:

* http://jdebp.uk./Softwares/nosh/

I get your sentiment, but creating a minimalist, but compatible alternative to systemd in a safe language is anything but reinventing the wheel. There's a real concern with systemd's feature creep, it's defaults, security and a development culture. rustysd may be a proof of concept, but it addresses real issues.

From the README:

> For now that is just out of interest how far I could come with this and what would be needed to get a somewhat working system. It is very much a proof of concept.

It sounds to me like this was more of a project that was just for fun and not really meant to be a genuine attempt at producing an alternative to systemd.

Some of the best and most lasting systems start out this way.

Yes, but you need to be an expert in NLP to write that NLP library. People in their free time write software that sits at the intersection of interesting and able.

Well said :)

On the one hand yes there's tons to explore, so many promising futures.

But I would like to keep an open mind about works like this one. This is work that could beget all kinds of neat things! Systemd targets Linux, where as this is a bit higher level & might be portable to other environments. It allows other people to not have to create their own process management systems, not have to create their own init system. The recreation brings the availability of the giant's shoulder to many many more places, distributes it's promise to really new & interesting targets.

It's pretty likely that the people writing an init system in Rust are not just gonna casually decide to write an advanced NLP library instead, since it requires a fairly distinct set of interests and skills.

Rust was created to replace C, not build anything radically new or different.

Technology advances by the continued reinvention of the wheel. As much as I dislike Rust the language, these activities of the Rust community are spot on.

Exactly, like the borrow checker. Ohh, wait.

This is the wheel of progress.

Reply on news.ycombinator.com