Newest at the top
| 2025-01-31 21:45:04 +0100 | <dminuoso> | So it gets woken up whenever the thunk finished. |
| 2025-01-31 21:44:48 +0100 | <dminuoso> | If another thread enters a blackhole, it gets put on a list to be woken up later. |
| 2025-01-31 21:44:33 +0100 | <dminuoso> | mauke: If the same thread enters a blackhole, that blackhole acts as loop detection., |
| 2025-01-31 21:44:25 +0100 | <dminuoso> | mauke: Okay, so there's two behaviors to blackhole. |
| 2025-01-31 21:44:08 +0100 | alexherbo2 | (~alexherbo@2a02-8440-3503-94e0-1866-04f2-f81a-c1ec.rev.sfr.net) (Remote host closed the connection) |
| 2025-01-31 21:44:07 +0100 | <mauke> | so if another thread tries to evaluate the same thunk later, it will simply wait until the first thread is done computing a value |
| 2025-01-31 21:43:33 +0100 | <mauke> | in a multi-threaded environment, the first thread to reach a given thunk instead switches out the code pointer to an "enter waiting queue" subroutine |
| 2025-01-31 21:42:48 +0100 | <mauke> | at least in single-threaded mode |
| 2025-01-31 21:42:30 +0100 | <mauke> | this is implemented by temporarily switching the code pointer of a thunk to a subroutine that throws an exception |
| 2025-01-31 21:41:58 +0100 | euouae | furiously types some notes of the previous discussion, needs more time to read the latest stuff being said |
| 2025-01-31 21:41:55 +0100 | <mauke> | anyway, the <<loop>> exception happens when evaluation of a thunk tries to re-enter the same thunk (i.e. you have a value that depends on itself) |
| 2025-01-31 21:41:06 +0100 | <mauke> | huh, interesting |
| 2025-01-31 21:40:03 +0100 | <dminuoso> | See https://simonmar.github.io/bib/papers/multiproc.pdf |
| 2025-01-31 21:39:37 +0100 | <mauke> | I may be wrong on my terminology, but I think blackholing only happens in single-threaded mode |
| 2025-01-31 21:39:25 +0100 | <dminuoso> | And that all alone pretty much stops memory issues. |
| 2025-01-31 21:38:58 +0100 | <dminuoso> | euouae: That is, there is automatic protection that no two threads attempt to evaluate the same expression concurrently. |
| 2025-01-31 21:38:38 +0100 | <lambdabot> | *Exception: <<loop>> |
| 2025-01-31 21:38:37 +0100 | <dminuoso> | euouae: First off, when switchinig between threads, entered thunks are blackholed. |
| 2025-01-31 21:38:36 +0100 | <mauke> | > let x = head [x] in x |
| 2025-01-31 21:38:18 +0100 | <euouae> | No |
| 2025-01-31 21:38:12 +0100 | <mauke> | on a related topic, have you ever seen the <<loop>> exception? |
| 2025-01-31 21:38:02 +0100 | <dminuoso> | No, its all build to handle that. |
| 2025-01-31 21:38:02 +0100 | <euouae> | I can see why memory can blow up then |
| 2025-01-31 21:37:49 +0100 | <euouae> | Oh it does? That can be bad |
| 2025-01-31 21:37:40 +0100 | sprotte24 | (~sprotte24@p200300d16f06b9001d5c2b08794be0ce.dip0.t-ipconnect.de) |
| 2025-01-31 21:37:34 +0100 | <mauke> | euouae: yes |
| 2025-01-31 21:37:15 +0100 | <mauke> | my point is that just as main() is a regular C function (has an address, can be called from inside the program, etc), so Haskell main is a regular IO () value and can be used as such |
| 2025-01-31 21:37:11 +0100 | <dminuoso> | Oh you're really diving deep now. |
| 2025-01-31 21:36:58 +0100 | <euouae> | so what about threads? is sharing happening across threads? |
| 2025-01-31 21:36:43 +0100 | <dminuoso> | euouae: Right. |
| 2025-01-31 21:36:41 +0100 | <mauke> | yes |
| 2025-01-31 21:36:36 +0100 | <euouae> | very similar to C etc |
| 2025-01-31 21:36:32 +0100 | <euouae> | Well GHC bootstraps some code before main to do the execution |
| 2025-01-31 21:36:20 +0100 | <mauke> | the runtime system grabs the value of main and runs the instructions you built |
| 2025-01-31 21:36:05 +0100 | <mauke> | main is just a constant |
| 2025-01-31 21:36:02 +0100 | <mauke> | I don't think that's the right way to think about it |
| 2025-01-31 21:35:51 +0100 | <dminuoso> | That is, main is the only thing (aside from some dark primitives) that can execute IO. |
| 2025-01-31 21:35:35 +0100 | <dminuoso> | Of course *executing* IO also happens at runtime and is done by main. |
| 2025-01-31 21:35:33 +0100 | <euouae> | dminuoso: I'm very prone to using the wrong terms all the time |
| 2025-01-31 21:35:25 +0100 | <dminuoso> | Yes, you can imagine this happens at runtime. |
| 2025-01-31 21:35:16 +0100 | <euouae> | Yeah right, evaluating |
| 2025-01-31 21:35:14 +0100 | <mauke> | in my example, if nothing is shared, the code can run in O(1) memory |
| 2025-01-31 21:35:14 +0100 | <dminuoso> | euouae: But yes, you can imagine that substitution happens at runtime. |
| 2025-01-31 21:34:58 +0100 | <euouae> | But feigning ignorance on compilation, it's fine to think of it as all interpreted at runtime right? |
| 2025-01-31 21:34:57 +0100 | <dminuoso> | euouae: You mean *evaluating* right? |
| 2025-01-31 21:34:38 +0100 | <euouae> | So when "executing" haskell code, theoretically you can imagine that all the grammatical expansions happen at runtime. Obviously "compiled" haskell code can do some of the work before hand |
| 2025-01-31 21:34:38 +0100 | <lambdabot> | 19999 |
| 2025-01-31 21:34:36 +0100 | <mauke> | > let { x = [0 ..]; y = [0 ..] } in (x !! 10000) + (y !! 9999) |
| 2025-01-31 21:34:20 +0100 | <dminuoso> | But not sharing can also waste memory in potentially having to keep it in memory multiple times.. |
| 2025-01-31 21:34:08 +0100 | <euouae> | Ah yeah that makes sense |