[CM] Worthwhile to explore realtime-friendly GC strategies for s7?

[Kjetil Matheussen]

On Sat, Apr 10, 2021 at 1:41 AM Iain Duncan <iainduncanlists at gmail.com> wrote:
>
> Hello Bill and colleagues, this is mostly a question of curiosity, though potentially also a question of "wonder what would make a good phd topic in a couple of years".  I'm getting pretty good results running complete s7 sequencing environments in Max so long as my latency setting is high enough to be able to handle some fairly large bursts. I run the gc automatically every 100 to 300 ms and that seems to work pretty well at taming the bursts, but they are still there, introducing constraints at what I can do with lower latency (but surprisingly capable to be honest!).
>
> I'm just curious to hear thoughts. Would trying to put a more real-time friendly GC strategy into s7 be possible, feasible, worthwhile, etc? I know next to nothing yet about the various ways GCs work, but I'm getting lots of support (from my supervisors at least) for working on Scheme for Max and Scheme for PD as worthwhile long term projects, and I'm definitely interested in looking under the hood more.  I'm not really interested in the no-GC approach used by things like Extempore and Carp, as the whole advantage for me is the high level dynamic coding, but if making the GC more real time friendly is not a fool's task, it could be something I might be able to work towards in the long run
>

This gc might be interesting:
http://users.notam02.no/~kjetism/rollendurchmesserzeitsammler/
Snd was even the main testbed for it.
Note that it was made before it was necessary for memory access to be
atomic for correct behavior, so it probably needs some modifications
to run on modern computers. It might not be that much work though, and
the information in the master thesis should still be valid:
http://users.notam02.no/~kjetism/rollendurchmesserzeitsammler/Matheussen_Master.pdf