00:58 <sthalik> hey guys

00:58 <sthalik> did you consider making pypy capable of saving itself to disk and resuming, much like Lisp VM's do?

00:58 <sthalik> that would solve the tracing problem

00:59 <sthalik> i.e. restore the core containing the stack and heap

01:00 <sthalik> lisp compilers even have fun stuff like marking all saved-to-VM heap allocations as permanent

01:00 <sthalik> that way, GC on large heaps is cut by a lot

01:00 <sthalik> IIRC only lisp and smalltalk do this; supposedly linux has some checkpointing tool...

09:59 <cfbolz> sthalik: yes, we have talked about this. but it's a very hard problem

10:00 <sthalik> does the difficulty also apply to ready-made checkpointing programs for linux?

10:02 <cfbolz> how do those deal with stuff like open files

10:04 <cfbolz> anyway, either you use something readymade (then you can already do that and don't need any pypy specifics). or you don't, and then it's a research project 😅

10:05 <nimaje> that only "solves" the problem for long running programs that you suspend and resume later, not for some calculations or stuff that you run with diffrent parameters, well you could model the interpreter as long running program that you suspend, but then it would keep all traces and have to decide when to get rid of them and where to keep the image?