July 17, 2025: Computer Programming Language Panel

[Serg Gini]

On Thursday, 7 August 2025 at 10:10:05 UTC, Bienlein wrote:
> On Thursday, 17 July 2025 at 08:30:01 UTC, Serg Gini wrote:
>> On Thursday, 17 July 2025 at 07:12:46 UTC, Walter Bright wrote:
>>> at 6pm I'll be representing D as a panelist.
>>>
>>> Details:
>>>
>>> https://nwcpp.org/
>>>
>>> "This month we will be hosting a joint meetup with the Rust 
>>> User Group to discuss a breadth of programming languages with 
>>> a panel of experts. In this talk we will compare and 
>>> contrast: D, C++, Rust, Swift, Cpp2, C# and Mojo."
>>
>> In case somebody will say that "GC is slow"
>> https://github.com/jinyus/related_post_gen
>
> I don't want to be negative, just in order to remain factual: I 
> don't see in what way this related_post_gen benchmark causes 
> load on the GC. Also, the results don't list memory usage which 
> renders it almost useless as a GC benchmark.

Hi there. This message didn't mean that this benchmark is a 
proper GC load.
But it is the response to some group of people that are saying 
"This language has GC - it's slow. I don't want to use it".
So the point was that even the language with GC can be fast. If 
you use proper memory management techniques or in some tasks 
where memory is less important.

> The binarytrees benchmark on 
> https://programming-language-benchmarks.vercel.app/ is a good 
> test for the GC. I'm aware of the discussion that this site to 
> some extend does not compare the same things in their 
> implementation of the respective programming languages. But it 
> is quite striking to see that the D GC consumes a multiple of 
> the memory some other GCes language like Crystal consumes:
>
> https://programming-language-benchmarks.vercel.app/d-vs-crystal
>
> Same picture wehen you compare D with Go or Java:
>
> https://programming-language-benchmarks.vercel.app/d-vs-go
> https://programming-language-benchmarks.vercel.app/d-vs-java
>
> For the binarytrees benchmark you can see that the other 
> languages must have a more efficient GC (cpu time lower, memory 
> consumption lower ot both). In case of Java and Go that is 
> acceptable to me. Both languages are being backed by multi 
> billion companies and have very sophisticated GCs. But Crystal 
> is a hobbyist open source language much simpler than D and its 
> GC seem to perform much better.

Ok now about this code.
General rule - always compare code not only results.

D code for some slight improvements in performance is adding 
special rule to not grab garbage that lower than 300 Mb.
Without this GC setting the memory consumption will drop to 60-70 
Mb.

D's GC also has some other settings that can be tuned for better 
results.

Also different languages have different default conditions. For 
example, C# is allocating initially quite a big chunk of memory, 
which helps them to have better results on such GC-stressful 
tasks.

Also keep in mind, that this website is prohibiting direct use of 
Arenas, while some languages are using them internally.
All these tradeoffs of internal implementations can help them to 
have higher values in some tasks (like binarytrees problem), but 
at the same time show worse results in other tasks.

And last, but not least there are of course other ways to solve 
this problem.

https://github.com/BinaryTrees
You can find D solution that should outperform the one that 
available in programming-benchmark-game.
They are just not in scope of the comparison. But for real 
problem - you are not locked withing "standard GC solution"