| 12 Jan 2026 |
 Randy Eckenrode | I would like to try structuring the bootstrap as a cross (with prevStage really being buildPackages). It probably wouldn’t reduce the number of builds, but it might allow is to drop most of the overlays. | 18:15:02 |
| Randy Eckenrode | But that is like 26.11 or 27.05 stuff. | 18:15:28 |
| Randy Eckenrode | Aside from this cleanup, I want to switch Darwin to LLD by default. That will require some work because LLD is not wrapped properly for that. | 18:16:01 |
| Randy Eckenrode | No cctools or ld64 should be in the bootstrap at that point. | 18:16:51 |
 emily | hmm, do we actually use python3-bootstrap as our final pkgs.python3? | 18:17:36 |
| emily | does that mean Darwin gets Python without any LTO? | 18:17:41 |
| Randy Eckenrode | I also want to drop the source releases from the SDK and vendor the needed sources along with the others. | 18:18:00 |
| emily | or is it just that we use one python3-bootstrap build throughout the entire bootstrap but then it gets built once more for the real python3? | 18:18:13 |
| Randy Eckenrode | I hope not. That would be a mistake. | 18:18:16 |
| emily | right | 18:18:20 |
| emily | was just thinking if it makes sense to think about trimming some of the other feature flags again if it can save us significant builds | 18:18:35 |
| emily | since we pretty much just need it for Meson and the LLVM/Ninja builds | 18:19:24 |
| Randy Eckenrode | Not really worth the effort. LLVM and Clang dominate the build times on Darwin. Maybe only OpenSSL is as bad. | 18:19:42 |
| emily | fair enough | 18:19:54 |
| emily | I mean it's probably not that bad to pull in Lua or Tcl either ultimately | 18:20:08 |
| Randy Eckenrode | I would be open to switching to Swift Crypto for ld64, but if we can switch to LLD, it doesn’t matter. | 18:20:19 |
| emily | it does affect what things rebuild the world for staging though | 18:20:22 |
| Randy Eckenrode | That would be the only thing. Is sqlite really not a dependency in the bootstrap? | 18:20:45 |
| Randy Eckenrode | It totally will if Swift ever gets added. | 18:20:58 |
| emily | it's only used via libiconv → ATF → Kyua → SQLite and Python → SQLite | 18:21:11 |
| emily | Kyua pulls in Lua too, SQLite will pull in Tcl soon | 18:21:25 |
| Randy Eckenrode | I’m getting close on Swift. I need to fix it to find the separate stdlib, then I need to see if I can use prebuilts. | 18:22:45 |
| emily | if Swift pulls in SQLite regardless then maybe it's not worth trying to cut it out now | 18:23:06 |
| Randy Eckenrode | Then do Linux, but I think that will be okay with LLVM overrides being only somewhat broken now. | 18:23:07 |
| Randy Eckenrode | * | 18:23:26 |
| emily | are you at the point where you'd like broad architectural review for the Swift stuff btw? not sure what state that stuff has in, have been holding off on looking when there's no PR yet 😅 | 18:23:55 |
| emily | though of course my reliability at looking at stuff is less than I'd like in general lately… | 18:24:07 |
| emily | (will still try to look at that stdenv PR…) | 18:24:16 |
| Randy Eckenrode | I can push a branch where it’s just a big, mega commit. | 18:24:33 |
| Randy Eckenrode | I started over with my previous work as a base. | 18:25:03 |
