Garbage Collection for Systems Programmers (page 7)

April 06

Re: Garbage Collection for Systems Programmers

Posted by Carl Sturtivant
in reply to Sergey

Permalink

Carl Sturtivant

Posted in reply to Sergey

Permalink

On Saturday, 6 April 2024 at 17:08:55 UTC, Sergey wrote:

On Saturday, 6 April 2024 at 16:59:14 UTC, Carl Sturtivant wrote:

On Saturday, 6 April 2024 at 16:28:25 UTC, Sergey wrote:

So what is the point then you wanted to discuss in this thread?

Finding strategic purpose via the future of GC.

D doesn't have any strategy.

As discussed, it needs to get some.

On Saturday, 6 April 2024 at 00:29:04 UTC, H. S. Teoh wrote:

On Sat, Apr 06, 2024 at 12:06:04AM +0000, Adam Wilson via

This is the point I was trying to make. The strategic win for have a fantastic GC would be immense, and would far outweigh anything we could gain by continuing down the no-GC path.

Totally. If we could introduce write barriers that opens the door to incremental generational GCs, that would be a HUGE step in the long run. Maybe not so much in the short term, but guys, it's been at least a decade. We gotta stop chasing decisions that only benefit the short term. It's time to think about long-term strategy.

On Saturday, 6 April 2024 at 16:49:44 UTC, Richard (Rikki) Andrew Cattermole wrote: > Modern GC's tend to need write barriers to make them work. Write barrier can be as cheap as a single atomic write.

On Sat, Apr 06, 2024 at 07:09:12PM +0000, Sebastian Nibisz via Digitalmars-d wrote: > On Saturday, 6 April 2024 at 16:49:44 UTC, Richard (Rikki) Andrew Cattermole > wrote: > > Modern GC's tend to need write barriers to make them work. > > Write barrier can be as cheap as a single atomic write. So let's do it! T -- The problem with the world is that everybody else is stupid.

On Saturday, 6 April 2024 at 21:04:03 UTC, Ogi wrote:

On Saturday, 6 April 2024 at 14:53:09 UTC, Sergey wrote:

Very similar test..
D with LDC showed more FPS than Rust and C++ :P
https://github.com/NCrashed/asteroids-arena

This thing barely uses GC. It implements ECS architecture: all game entities are split into small components which are in separate arrays in a manner of an object pool pattern. And it utilizes std.container.Array to avoid GC.

Object pool may be good for performance, but’s also a form of manual memory management and therefore haram. GC bros want us to “just use GC”.

The fallacy is to push for a XOR strategy instead of a OR strategy in regards to resource management.

The GC bros don't argue GC is the only strategy, only that it should be the default.

On 07/04/2024 7:09 AM, Sebastian Nibisz wrote: > On Saturday, 6 April 2024 at 16:49:44 UTC, Richard (Rikki) Andrew Cattermole wrote: >> Modern GC's tend to need write barriers to make them work. > > Write barrier can be as cheap as a single atomic write. *can be* It would need to call into the GC as the strategy until we know what implementations need. It must also be noted that dmd-be doesn't support atomics as intrinsics and Walter has been very against adding it, so it would still be a function call.

On Sunday, 7 April 2024 at 06:16:13 UTC, Richard (Rikki) Andrew Cattermole wrote: > On 07/04/2024 7:09 AM, Sebastian Nibisz wrote: >> On Saturday, 6 April 2024 at 16:49:44 UTC, Richard (Rikki) Andrew Cattermole wrote: >>> Modern GC's tend to need write barriers to make them work. >> >> Write barrier can be as cheap as a single atomic write. > > *can be* > > It would need to call into the GC as the strategy until we know what implementations need. > > It must also be noted that dmd-be doesn't support atomics as intrinsics and Walter has been very against adding it, so it would still be a function call. Don't wait for Walter, you are capable, if you know how to do it, submit a PR I'm pretty sure everyone wants atomics as intrinsics, myself included, just do it

On Saturday, 6 April 2024 at 19:09:12 UTC, Sebastian Nibisz wrote: > On Saturday, 6 April 2024 at 16:49:44 UTC, Richard (Rikki) Andrew Cattermole wrote: >> Modern GC's tend to need write barriers to make them work. > > Write barrier can be as cheap as a single atomic write. Again, asking to become even worse than Java Java's latest new concurrent GC doesn't need write barriers Rust is the best evidence that nobody wants a system language with a GC, only Microsoft do, and they has lost

On 07/04/2024 9:41 PM, ryuukk_ wrote: > On Sunday, 7 April 2024 at 06:16:13 UTC, Richard (Rikki) Andrew Cattermole wrote: >> On 07/04/2024 7:09 AM, Sebastian Nibisz wrote: >>> On Saturday, 6 April 2024 at 16:49:44 UTC, Richard (Rikki) Andrew Cattermole wrote: >>>> Modern GC's tend to need write barriers to make them work. >>> >>> Write barrier can be as cheap as a single atomic write. >> >> *can be* >> >> It would need to call into the GC as the strategy until we know what implementations need. >> >> It must also be noted that dmd-be doesn't support atomics as intrinsics and Walter has been very against adding it, so it would still be a function call. > > Don't wait for Walter, you are capable, if you know how to do it, submit a PR > > I'm pretty sure everyone wants atomics as intrinsics, myself included, just do it I don't know how to do it and have zero desire to learn that backend to that point, let alone how to introduce such intrinsics. I know enough of x86 assembly to be dangerous, but over all have very little experience with backends. There are better ways for me to contribute.

Forums