libera/#clasp - IRC Chatlog

17:48:29 yitzi ping

17:48:29 Colleen yitzi: drmeister said 1 hour, 50 minutes ago: Are you running with both Apple Silicon and X86 homebrew installed at the same time? I'm exclusively X86 and I'm afraid to touch Apple Silicon while I'm actively developing.

17:49:04 yitzi drmeister: I have both installed

21:29:16 bike will unbound<Whatever_O>() always be the same thing? i'm trying to use it in a CAS

21:34:43 bike nevermind, it doesn't matter

22:50:23 yitzi llvm17 just got released. Homebrew has a pending PR for it.

1:14:09 bike got fdefns working on the runtime end, which is what i was least sure about. i called them "function cells" since that's slightly more descriptive

1:14:24 bike they're still stored in symbols, but that can be changed up later

3:09:55 drmeister yitzi: Awesome. I'm no longer afraid of Apple Silicon.

3:13:31 bike and fdefns working efficiently in bytecode now

3:13:36 bike cleavir i haven't really thought through yet

3:26:30 drmeister llvm python bindings have been removed. Apparently they haven't been maintained since llvm 10

3:31:27 bike we don't use them, do we?

3:34:50 drmeister Nope.

3:36:50 drmeister Wow, I just did something I think is amazing with Common Lisp/Cando for recognizing molecules. I have this subgraph matcher for matching large collections of subgraphs. Think regular expression like [abcd][abcd][efgh]. It will recognize "aae" "acg" and so on.

3:38:00 drmeister I wrote a method to recursively deep copy the subgraph matcher but change the node matchers to a thing called an `untrained-monomer-match-node`. That creates a matcher that's like ???

3:38:38 drmeister Then I give it a valid pattern to train it like "acg" and it mutates the subgraph matcher to something like [a][c][g].

3:39:56 drmeister It solves a nasty problem that I have matching overlapping subgraphs.

3:40:16 drmeister It depends on CHANGE-CLASS. I'm happy with it.

3:41:04 bike Cool

3:41:14 drmeister It's more complicated because it works with branched graphs where the edges have names that have to match.

3:42:18 drmeister It's one of those things where I think: "I have no idea how I would have done this thing in Python".

3:47:10 drmeister Do any of these llvm17 changes to DWARF impact us?

3:47:19 drmeister https://releases.llvm.org/17.0.1/docs/ReleaseNotes.html#changes-to-the-debug-info

3:47:31 drmeister The pass manager changes will impact us. That's going to take some work.

4:15:00 bike we use llvm.dbg.addr in dbg-variable-alloca, but all uses of it are dummied out right now, so i think we're good