Conversation

Notices

1. Embed this notice
   julesh (julesh@mathstodon.xyz)'s status on Wednesday, 08-Jan-2025 02:15:55 JST julesh

   It's likely I have absolutely no idea what I'm talking about here, but:

   If you're trying to compile a language with features like continuations to machine code, I wonder if it would be a good idea to ignore the cpu's stack, allocate your entire memory to heap and roll your own linked list structure to store your local variables. Function calls become significantly slower because you have to make a call to malloc to get a new frame, but the upside would be that reordering the stack becomes extremely cheap and no longer involves memcpy

   • Embed this notice
     VojtechStep (vojtechstep@mathstodon.xyz)'s status on Wednesday, 08-Jan-2025 02:21:46 JST VojtechStep

     in reply to

     @julesh that kinda sounds like what Chicken scheme does? Except with the twist that as an implementation detail, you also try to keep your heap on the stack as long as possible https://www.more-magic.net/posts/internals-gc.html

   • Embed this notice
     julesh (julesh@mathstodon.xyz)'s status on Wednesday, 08-Jan-2025 03:02:13 JST julesh

     in reply to

     I am pleasantly surprised by how much this isn't a bad idea

   • Embed this notice
     Bartosz Milewski (bartoszmilewski@mathstodon.xyz)'s status on Wednesday, 08-Jan-2025 03:13:32 JST Bartosz Milewski

     in reply to

     @julesh I think we are cursed with the ancient processor architectures tied up to stack machines. It seemed like a good idea back in the 20th century. Now you may have a computer with terabytes of memory, but you still blow your fixed-size stack on a recursive procedure.

   • Embed this notice
     psu_13 (psu_13@mathstodon.xyz)'s status on Wednesday, 08-Jan-2025 05:19:21 JST psu_13

     in reply to

     @julesh Didn't SML/NJ do essentially something like this? The "call stack" was heap allocated.

     <google>...

     Yeah something like this

     https://www.cs.princeton.edu/~appel/papers/stack2.pdf

   • Embed this notice
     Jencel Panic (abuseofnotation@mathstodon.xyz)'s status on Wednesday, 08-Jan-2025 17:08:07 JST Jencel Panic

     in reply to

     @julesh How do you use the stack at all, if your functions can be passed around (and their scopes, saved)?

   • Embed this notice
     Rhialto (rhialto@mathstodon.xyz)'s status on Monday, 20-Jan-2025 04:42:31 JST Rhialto

     in reply to

     @julesh You just reinvented the IBM 360/370/zArch calling convention.