Newest at the top
| 2025-05-08 11:44:19 +0200 | <hellwolf> | :D code is better than paper? |
| 2025-05-08 11:38:17 +0200 | Typedfern | (~Typedfern@213.red-83-37-26.dynamicip.rima-tde.net) typedfern |
| 2025-05-08 11:37:28 +0200 | Typedfern | (~Typedfern@213.red-83-37-26.dynamicip.rima-tde.net) (Remote host closed the connection) |
| 2025-05-08 11:36:44 +0200 | Typedfern | (~Typedfern@213.red-83-37-26.dynamicip.rima-tde.net) typedfern |
| 2025-05-08 11:32:46 +0200 | euleritian | (~euleritia@dynamic-176-006-143-223.176.6.pool.telefonica.de) |
| 2025-05-08 11:29:59 +0200 | euleritian | (~euleritia@ip4d17f864.dynamic.kabel-deutschland.de) (Ping timeout: 245 seconds) |
| 2025-05-08 11:29:43 +0200 | <tomsmeding> | but the documentation part never happened, lol |
| 2025-05-08 11:29:37 +0200 | <tomsmeding> | that's part of the reason I wanted to make an implementation: to then proceed to document that well so that it can serve as a better explanation for haskellers |
| 2025-05-08 11:29:14 +0200 | <tomsmeding> | I find the presentation in the McDonell paper to be hard to understand personally |
| 2025-05-08 11:28:34 +0200 | <hellwolf> | thanks, printed. will read to enrich my knowledge here. it seems relevant to what I do. |
| 2025-05-08 11:26:15 +0200 | <tomsmeding> | The McDonell et al paper presents an algorithm that does sharing recovery on an AST indexed by the type of the expression that it represents |
| 2025-05-08 11:25:46 +0200 | <tomsmeding> | but the expression it creates is untyped |
| 2025-05-08 11:25:38 +0200 | <tomsmeding> | the pitch is: https://ku-fpg.github.io/files/Gill-09-TypeSafeReification.pdf does sharing recovery just fine, and it's implemented in https://hackage.haskell.org/package/data-reify |
| 2025-05-08 11:25:13 +0200 | <tomsmeding> | hellwolf: yeah only the sharing recovery part of it |
| 2025-05-08 11:19:49 +0200 | rvalue- | rvalue |
| 2025-05-08 11:13:25 +0200 | rvalue | (~rvalue@user/rvalue) (Ping timeout: 248 seconds) |
| 2025-05-08 11:13:21 +0200 | fp | (~Thunderbi@2001:708:20:1406::10c5) fp |
| 2025-05-08 11:13:13 +0200 | rvalue- | (~rvalue@user/rvalue) rvalue |
| 2025-05-08 11:11:49 +0200 | j1n37- | (~j1n37@user/j1n37) (Ping timeout: 248 seconds) |
| 2025-05-08 11:11:32 +0200 | j1n37 | (~j1n37@user/j1n37) j1n37 |
| 2025-05-08 11:08:55 +0200 | <hellwolf> | | This is an attempt at implementing "Optimising Purely Functional GPU Programs" by McDonell et al. (ICFP 2013) in a generic fashion. |
| 2025-05-08 11:08:26 +0200 | <hellwolf> | let me read your project: one-liner README :) |
| 2025-05-08 11:07:42 +0200 | <tomsmeding> | because my partial-order thing doesn't (easily?) work |
| 2025-05-08 11:07:32 +0200 | <tomsmeding> | but in your case it may not solve anything because you require the linear types for versioning too |
| 2025-05-08 11:06:52 +0200 | <tomsmeding> | (the core function there is sharingRecovery at the bottom of this file https://git.tomsmeding.com/sharing-recovery/tree/src/Data/Expr/SharingRecovery/Internal.hs#n454 ) |
| 2025-05-08 11:05:17 +0200 | <hellwolf> | I am all ears to hear feedback and criticism on my whole LVM approach |
| 2025-05-08 11:05:06 +0200 | <tomsmeding> | I'd like someone to use https://git.tomsmeding.com/sharing-recovery/about/ in anger :p |
| 2025-05-08 11:04:43 +0200 | <hellwolf> | (we talk about variable labeling another time, when I get to that; I am okay with what I have now for that.) |
| 2025-05-08 11:04:42 +0200 | <tomsmeding> | the unsafePerformIO ID generation trick is sucessfully implemented by horde-ad (a library a collaborator of mine is working on) |
| 2025-05-08 11:04:20 +0200 | <hellwolf> | ah, okay, I didn't fully understand that |
| 2025-05-08 11:04:20 +0200 | <tomsmeding> | variable labeling is implemented successfully by accelerate |
| 2025-05-08 11:04:04 +0200 | <tomsmeding> | no, the "use (v, exists w. w) instead of (v, v+1)" |
| 2025-05-08 11:03:56 +0200 | <hellwolf> | using unsafe io? |
| 2025-05-08 11:03:48 +0200 | <hellwolf> | your idea of the variable labeling? |
| 2025-05-08 11:03:42 +0200 | <tomsmeding> | and using the continuation monad to achieve said CPS is insufficient because we're doing it not for the runtime CPS behaviour, but for being able to _type-restrict_ the continuation |
| 2025-05-08 11:02:51 +0200 | <tomsmeding> | you end up having to CPS all code to express what I described |
| 2025-05-08 11:02:15 +0200 | <tomsmeding> | hellwolf: I think my idea doesn't work when embedding in haskell because an effectful operation does not have access to the continuation to impose type restrictions on; it's (>>=) that does, but (>>=) is separate form the effectful primitive operation |
| 2025-05-08 11:00:56 +0200 | yoneda | (~mike@193.206.102.122) |
| 2025-05-08 11:00:49 +0200 | j1n37 | (~j1n37@user/j1n37) (Ping timeout: 276 seconds) |
| 2025-05-08 10:59:02 +0200 | j1n37- | (~j1n37@user/j1n37) j1n37 |
| 2025-05-08 10:57:21 +0200 | <tomsmeding> | perhaps GHC is fine, I dunno |
| 2025-05-08 10:57:02 +0200 | <Lears> | How would GHC fail to cope with too much RankNTypes? Bluefin also requires polymorphic continuations everywhere, and it does just fine. |
| 2025-05-08 10:52:19 +0200 | <hellwolf> | I have no time to write down everything yet. |
| 2025-05-08 10:52:12 +0200 | <hellwolf> | thanks for being the rubber duck with me. you are the first one actually I conversed in details with what I built :D |
| 2025-05-08 10:51:55 +0200 | <tomsmeding> | https://git.tomsmeding.com/yahb2/tree/bwrap-files/start.sh |
| 2025-05-08 10:51:38 +0200 | <tomsmeding> | :) |
| 2025-05-08 10:51:31 +0200 | <hellwolf> | you thinki so, did it enforce it in the type system? |
| 2025-05-08 10:51:13 +0200 | <tomsmeding> | you have write access only to a small tmpfs |
| 2025-05-08 10:50:51 +0200 | <yahb2> | <interactive>:99:1: error: [GHC-76037] ; Not in scope: ‘System.Procerss.systme’ ; Note: No module named ‘System.Procerss’ is imported. |
| 2025-05-08 10:50:51 +0200 | <hellwolf> | % System.Procerss.systme "rm -rf /hahaha" |