Garbage Collection for Systems Programmers

[Lance Bachmeier]

On Tuesday, 9 April 2024 at 15:02:52 UTC, jmh530 wrote:
> On Tuesday, 9 April 2024 at 14:14:08 UTC, H. S. Teoh wrote:
>> [snip]
>>> 
>>> Sing along with me: ~~ write barriers ~~
>>> 
>>> :)
>>
>> Write barriers!  We've been singing this tune for literally 
>> years. It's about time we actually did it instead of merely 
>> talking about it!
>>
>>
>> T
>
> If something hasn't happened for years, usually there are good 
> reasons why. Walter raised some recently here [1].
>
> If you force everything to go through the GC, then you avoid 
> some complexity, but you have the added cost of write barriers 
> all the time.
>
> If you're in a mixed memory environment, you end up needing two 
> types of pointers (one managed, one unmanaged) and that leads 
> to a lot of complications that people tend to not want to deal 
> with.
>
> [1] https://forum.dlang.org/post/uv1l7p$2tu2$1@digitalmars.com

I honestly don’t find that line of argument very convincing. 
There are no numbers backing up the conclusions, and not even a 
specific application given. We already have experience with the 
status quo and the problems it can cause. There’s no 
justification for ruling out other options based on speculation 
that the performance hit will be unacceptable.