• james (2/2) May 24 Hello All,
• mike_jack (42/42) May 24 This is actually a thoughtful observation, and brings up the
• Walter Bright (6/9) May 26 D allows complete freedom in selecting memory allocation strategies. The...

james <marketing tier1app.com> writes:

Hello All,
Can you share thoughts on a Future Garbage Collector

mike_jack <rajas2647 gmail.com> writes:

This is actually a thoughtful observation, and brings up the 
common perspective many developers have after working extensively 
with garbage-collected languages like Java. In fact, your 
suggestion to move away from a monolithic, “one-size-fits-all” 
garbage collector and toward a more flexible memory management 
system aligns well with trends in systems-level language design.

 From your Java experience, it’s clear that GC behavior often 
becomes a bottleneck when performance is critical. With that, the 
developers starts slowly to focus more on GC, favoring arrays, 
object pooling, or stack allocations, in order to minimize the 
pauses or delays.  That’s why your suggestion for D to treat 
memory management as a first-class citizen in the language and 
standard library is very compelling.

Giving developers the ability to hint object lifetimes, force or 
schedule GC with timing constraints, or even plug in custom GC 
algorithms opens up powerful optimization opportunities. The idea 
of supporting multiple memory managers, each tailored for 
different workloads (e.g., short-lived transactional data vs. 
long-lived singletons), would bring D closer to giving 
fine-grained control like you’d see in C++, while still offering 
the safety and convenience of a GC when needed.

Giving developers the ability to hint object lifetimes, force or 
schedule GC with timing constraints, or even plug in custom GC 
algorithms opens up powerful optimization opportunities. The idea 
of supporting multiple memory managers, each tailored for 
different workloads (e.g., short-lived transactional data vs. 
long-lived singletons), would bring D closer to giving 
fine-grained control like you’d see in C++, while still offering 
the safety and convenience of a GC when needed.

This “open memory model” approach may really provide a 
feasibility for the D community to experiment and innovate. 
Without waiting for one universal GC to get improved, D can  
evolve to host a suite of GCs and memory managers. These are both 
optimized for different scenarios, something even Java has been 
slowly moving toward with G1, ZGC, and Shenandoah.

To get a clearer idea of how traditional garbage collectors like 
Java’s operate, including why long pauses happen and what 
triggers them, this blog on Java Garbage 
Collection(https://blog.gceasy.io/what-is-java-garbage-collection/) does a
great job explaining the mechanisms and trade-offs involved. It highlights
exactly the sort of constraints developers run into, which makes a case for
more flexible memory management approaches.

In short, rather than pursuing a perfect GC, giving developers 
the tools and hooks to manage memory as needed might be the most 
pragmatic and powerful path forward for D.

Walter Bright <newshound2 digitalmars.com> writes:

On 5/24/2025 6:37 AM, mike_jack wrote:
 In short, rather than pursuing a perfect GC, giving developers the tools and 
 hooks to manage memory as needed might be the most pragmatic and powerful path 
 forward for D.

D allows complete freedom in selecting memory allocation strategies. The dmd 
compiler itself uses all of them:

1. GC

2. stack allocation

3. malloc/free (and mechanisms based on them)