| 29 Nov 2021 |
 dzaima | thus, if there's no REPL, the GC just doesn't happen | 22:27:23 |
 ashermancinelli | yeah that's not that different from just letting the OS free whatever you leaked in the case of running a script then, yeah? | 22:27:24 |
| ashermancinelli | well that makes me feel a little better haha | 22:27:50 |
| dzaima | from what I understood, there's some magic CC<whatever> from a library that rowan uses for automatic GCing of references. i have no clue what magic makes it possible | 22:29:46 |
| ashermancinelli | hm. well you'd still have to help it mark things I'd imagine, but maybe the magic is just in the cycle detection? | 22:31:19 |
| dzaima | also it's got bacon in the name, so you can't not use it | 22:35:32 |
| dzaima | can't tell if that thing even can free cycles that don't involve weak pointers actually. But this seems more proper-ish - it seems to maybe somehow differentiate being stored in a structure vs being a top-level reference or something? (which is the proper thing to do, it's just that that's hard if you can't just detect that) | 22:43:51 |
 rowan | i think it should be able to but i havent attempted gc yet. i think mark/trace is a much easier way to start then progress to something like what golang does | 22:49:41 |
| rowan | the paper is http://researcher.watson.ibm.com/researcher/files/us-bacon/Bacon01Concurrent.pdf | 22:50:56 |
| dzaima | that's a generic GC paper. The important question, orthogonal to the paper, is retrieving the list of roots | 22:52:07 |
| rowan | this is what i did in erlang vm
i think it worked out alright but its been a while
https://github.com/cannadayr/ebqn/blob/main/src/ebqn_gc.erl#L69 | 22:53:50 |
| ashermancinelli | well if that's semistandard, then CXBQN has a much better startup time:
$ time ./BQN -e '2+2'
real 0m0.140s
user 0m0.098s
sys 0m0.042s
| 23:15:47 |
 Marshall | In reply to @ashermancinelli:matrix.org
Video on the BQN VM/runtime/compiler in case you're only on matrix Great work on the visualization here! Added a link to the VM page. | 23:22:34 |
| ashermancinelli | Hey thanks! | 23:22:48 |
| ashermancinelli | Hopefully I didn't say/show anything too misleading. | 23:23:46 |
| Marshall | Well I did just change the JS line count from "about 500" to "about 600", but 667 still is about 500. | 23:25:18 |
| Marshall | Also documented NOTM. It's in a weird spot because it's basically a special case of VFYM, which I classified as a header-related opcode, even though NOTM isn't header-specific. | 23:26:14 |
| ashermancinelli | It's ~500 when compared to the number I assumed it would be when I first started learning about the VM, and assumed it would be closer to lua's | 23:26:22 |
| Marshall | Counting things like the random number generator utilities is kind of silly anyway. | 23:28:00 |
| ashermancinelli | Yeah, I mean that vm animation is actually a dummed down version of the actual vm in js, so you could swap another comp unit in there and those slides/animations would still work. It's gotta be pretty simple for something like that. | 23:29:22 |
| dzaima | bytecode tests still have an old name for NOTM | 23:29:25 |
| Marshall | Yeah I think I gave that one up before I even got an implementation working. | 23:32:32 |
| ashermancinelli | I think it would be kinda fun to jump on a call/do a livestream where we hack together a vm as a community | 23:33:12 |
| Marshall | That could work. | 23:34:27 |
| ashermancinelli | Might make the language seem more accessible to folks | 23:35:01 |
| Marshall | I guess you get a good feel for the array model, but the interesting primitives do stay locked away in r1, until you get to implement those. | 23:37:37 |
| Marshall | I haven't explicitly said this yet, so 1.0 status report: features here are mostly marked with their fates now. | 23:41:06 |
| Marshall | I think I'll remove modifier partial application and returns from the spec now as I don't know that I want them in the language, that is, they might not be useful enough to justify the complexity. | 23:42:00 |
| ashermancinelli | Hey! How exciting! | 23:43:15 |
| Marshall | High-rank [] syntax is the only major feature missing now. Other than that there's some spec writing for already-implemented features, and the tricky Under cases like ⌾∾. I'll take a stab at those and either get them supported or make them optional in the spec (and hopefully finish them later). | 23:44:20 |
