Yeah, gonna be interesting. Software companies working on consumer software often don’t need to care, because:
- They don’t need to buy the RAM that they’re filling up.
- They’re not the only culprit on your PC.
- Consumers don’t understand how RAM works nearly as well as they understand fuel.
- And even when consumers understand that an application is using too much, they may not be able to switch to an alternative either way, see for example the many chat applications written in Electron, none of which are interoperable.
I can see somewhat of a shift happening for software that companies develop for themselves, though. At $DAYJOB, we have an application written in Rust and you can practically see the dollar signs lighting up in the eyes of management when you tell them “just get the cheapest device to run it on” and “it’s hardly going to incur cloud hosting costs”.
Obviously this alone rarely leads to management deciding to rewrite an application/service in a more efficient language, but it certainly makes them more open to devs wanting to use these languages. Well, and who knows what happens, if the prices for Raspberry Pis and cloud hosting and such end up skyrocketing similarly.Writing in Rust or “an efficient language” does nothing for ram bloat. The problem is using 3rd party libraries and frameworks. For example a JavaScript interpreter uses around 400k. The JavaScript problem is developers importing a 1GB library to compare a string.
You’d have the same bloat if you wrote in assembly.
Maybe you’re confusing memory (RAM) vs storage ? Because I converted some backend processing services from nodejs to rust, and it’s almost laughable how little RAM the rust counterparts used.
Just running a nodejs service took a couple of hundred mb of ram, iirc. While the rust services could run at below 10mb.
But I’m guessing that if you went the route of compiling an actual binary from the nodejs service, you could achieve some saving of ram & storage either way. With bun or deno ?
Because I converted some backend processing services from nodejs to rust,
You converted only the functions you needed and only included the functions you needed. You did not convert the entire node.js codebase and then include the entire library. That’s the problem I’m describing. A few years ago I toyed with javascript to make a LCARS style wall home automation panel. The overhead of what other people had published was absurd. I did what you did. I took out only the functions I needed, rewrote them, and reduced my program from gigabytes to megabytes even though it was still all Javascript.
Yeah, you need to do tree-shaking with JavaScript to get rid of unused library code: https://developer.mozilla.org/en-US/docs/Glossary/Tree_shaking
I would expect larger corporate projects to do so. It is something that one needs to know about and configure, but if one senior webdev works on a project, they’ll set it up pretty quickly.
On the one hand, tree shaking is often not used, even in large corporate projects.
On the other hand, tree shaking is much less effective than what a good compiler does. Tree shaking only works on a per-module basis, while compilers can optimize down to a code-line basis. Unused functions are not included, and not even variables that can be optimized out are included.
But the biggest issue (and one that tree shaking can also not really help against) is that due to the weak standard library of JS a ton of very simple things are implemented in lots of different ways. It’s not uncommon for a decently sized project (including all of the dependencies) to contain a dozen or so implementations of a padding function or some other small helper functions.
And since all of them are used somewhere in the dependency tree, none of them can be optimized out.
That’s not really a problem of the runtime or the language itself, but since the language and its environment are quite tightly coupled, it is a big problem when developing on JS.
Add to the list: doing native development most often means doing it twice. Native apps are better in pretty much every metric, but rarely are they so much better that management decides it’s worth doing the same work multiple times.
If you do native, you usually need a web version, Android, iOS, and if you are lucky you can develop Windows/Linux/Mac only once and only have to take the variation between them into account.
Do the same in Electon and a single reactive web version works for everything. It’s hard to justify multiple app development teams if a single one suffices too.
This equation might change a bit as more software users learn how bloated apps affect their hardware upgrade frequency & costs over time. The RAM drought brings new incentive to teach and act on that knowledge.
Management might be a bit easier to convince when made to realize that efficiency translates to more customers, while bloat translates to fewer. In some cases, developing a native app might even mean gaining traction in a new market.
We have an enormous problem with software optimization both in cycles and memory costs. I would love for that to change but the vast majority of customers don’t care. It’s painful to think about but most don’t care as long as it works “good enough” which is a nebulous measure that management can use to lie to shareholders.
Even mentioning that we’ve wiped out roughly a decade in hardware gains with how bloated and slow our software is doesn’t move the needle. All of the younger devs in our teams truly see no issue. They consider nextjs apps to be instant. Their term, not me putting words in their mouths. VSCode is blazingly fast in their eyes.
We’ve let the problem slide so long that we have a whole generation of upcoming devs that don’t even see a problem let alone care about it. Anyone who mentors devs should really hammer this home and maybe together we can all start shifting that apathy.
Everyone better start learning Rust.
Rust programs can definitely still consume a lot of memory. Not using a garbage collector certainly helps with memory usage, but it’s not going to change it from gigabytes to kilobytes. That requires completely rethinking how things are done.
That said I’m very much in favour of everyone learning Rust, as it’s a great language - but for other reasons than memory usage :)
True, but memory will be freed in a more timely manner and memory leaks probably won’t happen.
Memory leaks are more than possible in rust. Rust type system prevents things like free being called on an already free resource. It very much also allows not calling free even when nothing references things. It also makes things like arena allocation a fun endeavor compared to other systems languages. It’s not impossible just trickier. Rust isn’t a panacea, you would need something more like idris with its type system to programmatically enforce resources are freed at runtime during the compilation phase. But a fully dependent type system is very much a bleeding edge thing.
