Hacker News Reader

Comment by dataflow | Hacker News Reader

D never took off IMO because you can't abstract away a GC. The moment your callees rely on a GC, you have no choice but to have one in your program. The idea that you can avoid it in practice just because the language doesn't mandate it is therefore an illusion, and so it's not surprising at all that it isn't replacing C++.

View on HN

This is the logic I personally used in deciding not to use D.

Yeah, I'm pretty sure it's the reason many C++ folks reject it. It's kind of ironic that having the "wrong" useful feature can hurt you, but that seems to be what's happening here. I feel like they might actually have a shot at solving it by literally dropping GC support, but it's hard to say at this point. I don't know if the standard library still has anything that needs a GC, but making sure that's not the case might be a good first step if they're ever interested in going in this direction.

Conveniences like closures depend on gc

That's not true

https://dlang.org/spec/function.html#function-pointers-deleg...

C++ has had closures without a GC for over a decade now.

As syntactic sugar for how C++ functors used to be implemented, and they also bring their own set of issues when interoperating with co-routines.

If anything that's a statement about coroutines, not closures. And even then, not every coroutine has problems. I don't see a logical implication from that to a GC requirement.

> This is the logic I personally used in deciding not to use D.

Good call - I was unfortunate enough that I had to use it professionally for a few years. As comic book guy in the Simpsons would say "Worst language ever!".

Can you elaborate on your experience? There's a lot of negative D opinions from people who've barely used it and, for obvious self-selection reasons, a lot of positive ones from long term true believers. It would be interesting to see a negative experience from someone who used it extensively nonetheless.

I’m interested in hearing more. Also, I’m curious what language you use as a your daily driver?

I mean, the same is true for C++. If some C++ library you use employs GC, you can’t abstract that away either. The actual difference is how frequent such libraries are in the respective ecosystem.

> The actual difference is how frequent such libraries are in the respective ecosystem.

That’s a difference without a distinction: people will use the defaults unless they can’t.

So in C++ libraries will avoid GCs unless their work basically requires it, whereas in D they’ll require it unless the author is constrained to a nogc requirement.

I basically said as much https://news.ycombinator.com/item?id=34850130

Though, I'm not sure you could even have a precise GC for C++ even if you wanted to, given the way most C++ is written.

I'm not sure I follow

Is this about D code that interops with C/C++?

You have "extern (C++)" and @nogc for that

No, it's about pure D code. Some (many) people just don't want a GC in their program and that's why they use C++. If they switch to D, they'll have to put up with a GC if any of their dependencies needs one, and the probability of that being the case is... high.

@nogc eliminates that risk.

The GC also doesn't run if you don't call it so mitigating it's downsides while keeping the good aspects isn't all that hard.

You disable GC for the whole program? And what do you do when one of your dependencies relies on it? Just live with a memory leak?

Bear in mind you're simply not going to convince anti-GC folks that you can have the "good aspects" of a GC without the bad ones, and they're the folks you're trying to market to. To many C++ programmers the only good GC is one that is absent. You can tell them they're wrong all you want, but regardless of who's right, it's always been a losing battle.

It won't compile.

You can also make it so it will compile but won't collect unless you ask it to explicitly.

We don't exclusively market to C++ developers. D replaced PHP and Java for quite a few D users.

It won't compile, and... how is that a solution? You've basically told some users they can't use your language in that case. That's problem solved?

If you see PHP and Java replacement as taking off, that's awesome. I was addressing those who want to see it replace C++.

The express purpose of the @nogc attribute is to enforce that code under it's influence doesn't use the GC.

There is lots of @nogc code available to depend upon

> There is lots of @nogc code available to depend upon

That's not sufficient here. You can literally have a billion @nogc libraries, and even if there was one single GC-requiring library that many people want to use, the problem would be still there.

The only way to make this work is to ensure all of the libraries most people want to use work fine without a GC (or have @nogc alternatives that people would actually want to use). While not theoretically impossible, practically speaking, that's incredibly hard to pull off when you have a shiny GC sitting there.

In practice, "lots" isn't good enough. The whole ecosystem should work, not just lots.

In practice (I.e. actual practice that pays my rent) it's been good enough for me

Then why did D get lapped by Rust if @nogc is good enough and the ecosystem is good enough?

If the answer is along the lines of "corporate backing", why did Rust get corporate backing and D didn't?

Rust did memory safety better than D did, which was a very attractive prospect and then Rust was also lucky enough to get a following of people who very noisy about it.

Also in a world where JavaScript (excel, even) is dominant let's not ascribe these things to purely rational factors.

Aye, you can turn off GC entirely, for the whole program

There's even a mode of D that operates this way by default, "-betterC" meant to be a C replacement

The GC really is an entirety opt-in component.

I'd encourage you/other folks to give D an honest try, you might be surprised

You're not understanding my point unfortunately. I never said you can't turn it off. I said if you do, and any of your dependencies expect it to be on, then you have a problem. (Yes, I have tried D. That's how I know this.)

Ohhh, fair enough, yeah I see what you're saying.

Yeah you're boned in that case I think and that's a real problem that can pop up if you want full @nogc

EDIT: Maybe not based on Max's comment

D-Plug (Audio FX plugins) is one example of having to build up a dependency stack of @nogc compatible deps, a lot of that is in-house stuff.

Not impossible but it's a commitment, at which point you may as well use something else unless you REALLY want to use D, for sure.

> a lot of that is in-house stuff.

That's exactly the issue. Having to reinvent every wheel you need just isn't a selling point.

If you actually turn it off (as in GC.disable()) they won't know either way, it's just a memory allocator to them.

Reply on news.ycombinator.com