| 28 Jan 2026 |
 Ihar Hrachyshka | I think yes, parsing is the way. which is why I was saying - maybe worth instructing an agent with a sample to look for and let it do the job. I'm actually thinking I should take this shower thought seriously and build a tool. will call it 'aisect' :D | 18:24:12 |
 samasaur | that would be cool | 18:24:54 |
| samasaur | my bisect had been running fine but i just tried to SSH into that machine to check and apparently it went offline 4 minutes ago | 18:25:20 |
| Ihar Hrachyshka | the agent could also do the doCheck = true -> false flips I think etc. | 18:25:28 |
| samasaur | which i fear may have been my fault | 18:25:33 |
| Ihar Hrachyshka | I hit bison failures from get go on first good-to-bad iteration | 18:26:04 |
| Ihar Hrachyshka | * I hit bison test suite failures from get go on first good-to-bad iteration | 18:26:12 |
| samasaur | i checked an hour ago or so and the bisect was still running, though ofc it could have logged a bad rev due to deps | 18:26:57 |
 Randy Eckenrode | Hmm. D-Bus 1.16 appears to expect that it will be socket-activated with launchd on macOS. | 18:47:25 |
| Randy Eckenrode | That’s a problem for packages that need D-Bus during their build/tests. | 18:47:51 |
| samasaur | ah ha my build machine is back online! and the bisect finished! unfortunately it has pointed me to d4fd918e which has changes that are cygwin-only, so im gonna take a wild guess and say that it is not the commit causing the problem | 18:49:28 |
| Ihar Hrachyshka | I'm still early in the process, at a point where every step rebuilds the world (as you shrink the diff, the chance is higher that the next step won't require a world rebuild). let's see who's first. :p | 18:52:32 |
| Ihar Hrachyshka | swift> Running phase: unpackPhase
swift> unpackPhase completed in 28 minutes 11 seconds
is this... normal?
| 22:06:48 |
 xored | swift strikes again | 22:13:30 |
| Ihar Hrachyshka | I think it may be my old with the node where coreutils for some reason starts lagging | 22:15:03 |
| Randy Eckenrode | It’s not fast. Swift unpacks itself, LLVM, etc. | 22:15:12 |
| Ihar Hrachyshka | I see xargs ... sed now in patchPhase sitting for 10 mins+ patching | 22:15:21 |
| Ihar Hrachyshka | I had it in the past where at random moments builds stalled. maybe the issue is back. | 22:15:52 |
| Ihar Hrachyshka | lots of these in dmesg
[1476558.276431]: iterate_dirents_case_or_norm_insensitive:438: disk3s7 dirent hash chain iteration, dir_id 6242, hash 0x1c0f2e, count 20 [1476558.286126]: iterate_dirents_case_or_norm_insensitive:438: disk3s7 dirent hash chain iteration, dir_id 6242, hash 0x1c1a4c, count 20
| 22:19:11 |
| Ihar Hrachyshka | as per the Expert (chatbot), "You have one or more directories with a huge number of entries that collide badly under APFS’s case-insensitive hashing.
Every lookup becomes O(n) instead of O(1), and the filesystem gets slow." wonder if it's tripping or onto smth | 22:19:54 |
| Ihar Hrachyshka | ok after ~30 mins unpacking and ~15 mins of churning through the sed command finally it's building... jeez | 22:23:58 |
| samasaur | that does unfortunately sound like a typical swift build | 22:35:58 |
| Randy Eckenrode | The new Swift stuff is much faster, but that doesn’t help here. | 22:37:07 |
| samasaur | I imagine just taking the libcxx and LLVM out of the One Big Derivation helps a lot | 22:37:49 |
| Ihar Hrachyshka | nice. wonder if something can be done with it on APFS side, as the Expert suggested to use case-sensitive mode for nix store. | 22:37:56 |
| Randy Eckenrode | Some parts can also be done in parallel, and the default phase implementations are way faster at applying patches. | 22:39:04 |
| Ihar Hrachyshka | hm. I have some dyld sigsegvs in Console.app. maybe related to failed swift runs with the crash. was it dyld crashing? | 22:49:10 |
| Ihar Hrachyshka | nah it was swift-frontend (I also have some of these) | 22:52:43 |
| Ihar Hrachyshka | btw on a different machine...
swift> Running phase: unpackPhase
swift> unpackPhase completed in 30 seconds
| 22:54:50 |
| Ihar Hrachyshka | so no, this is not normal that it takes 30 mins. Maybe it's COMMON that folks struggle over slow APFS nix store, but it's not the only way it can be | 22:55:54 |
