Is there any roadmap for improving GC?

[ygc369]

Are there any new GC algorithms in developing?
Would reference counting collector be an alternative memory management mechanism?

[swgillespie]

Linking threads: dotnet/csharplang#547

[swgillespie]

What advantages do you think are to be had by switching to a reference counting collector instead of improving the GC we already have? Tracing collectors and reference counting collectors are duals of one another, and each has their own tradeoffs.

cc @Maoni0

[ygc369]

@swgillespie
Except RC and Local GC, what can be done to improve the GC we already have? Is anybody doing this?

[mikedn]

what can be done to improve the GC we already have?

But what exactly do you want to improve about GC?

You keep creating all this GC related issues but you never provide any sort of details - use cases, expectations, numbers, anything - and thus not exactly constructive.

[ygc369]

@mikedn
I'm sorry. But my mother tongue is not English. It is not easy for me to write a long English article with many details. I can only describe my ideas and I think their advantages are obvious and do not need more explainations.

[mikedn]

I think their advantages are obvious and do not need more explainations

Nope, there is nothing obvious about the advantages of reference counting.

[jkotas]

There have been thousands of research papers written about garbage collection, many of them on reference counting. We are aware of all the key ideas that have been published to date. There are folks in Microsoft Research working in this space too. It is very easy to pick and recommend an (existing) idea. The very hard part is how to incorporate it to the system to make it work better for significant fraction of important scenarios.

It is well established that simple reference counting does not work well because of the high cost of the reference count maintenance and inability to collect cycles. There are modifications that try to address it, but nothing obviously better accross the board.

http://gchandbook.org/ is a great summary of different garbage collection ideas with discussion of its process and cons. I would recommend that you to read this book, and try to implement and measure one or two ideas from this book to get a feel for how memory management works in practice before submitting more suggestions here. If you are not able to get access to the book where you live, send me email to my Microsoft address and we will figure out how to get you one.

[jkotas]

Duplicate of dotnet/csharplang#547 and #4029