1 00:00:00,120 --> 00:00:05,260 So, uh, if you tried to fire up Codex Code Mode last week on a Mac right after the 2 00:00:05,300 --> 00:00:11,070 v0.144.0 release, you- you- you probably got hit with an immediate, 3 00:00:11,380 --> 00:00:12,840 unceremonious crash. 4 00:00:13,460 --> 00:00:13,690 Just... 5 00:00:14,100 --> 00:00:15,530 boom, dead on launch. 6 00:00:16,200 --> 00:00:19,670 And thanks to Jellypod for keeping us sponsored while we dissect this, 7 00:00:20,220 --> 00:00:24,540 because the underlying architectural shift here is actually massive. 8 00:00:25,240 --> 00:00:32,200 It- it- it's all about this new background process called codex-code-mode-host. 9 00:00:33,748 --> 00:00:36,508 Right, the codex-code-mode-host. 10 00:00:37,088 --> 00:00:41,728 It sounds like a ghost process, but it's actually a sibling process model, 11 00:00:41,908 --> 00:00:42,608 right? 12 00:00:42,628 --> 00:00:47,588 They're splitting the heavy lifting away from the main Terminal User Interface. 13 00:00:47,592 --> 00:00:48,692 Exactly. 14 00:00:48,752 --> 00:00:52,202 Before this, the main TUI was trying to handle the UI rendering, 15 00:00:52,872 --> 00:00:58,492 manage the intensive compilation tasks, and run these deep agentic loops all in a 16 00:00:58,532 --> 00:00:59,182 single thread. 17 00:00:59,852 --> 00:01:03,932 And what happens when an agent loop gets stuck parsing a massive AST? 18 00:01:04,572 --> 00:01:05,652 The UI lags. 19 00:01:06,172 --> 00:01:07,532 It- it- it stutters. 20 00:01:08,072 --> 00:01:11,812 So, they offloaded the runtime loops entirely to this new host binary. 21 00:01:12,416 --> 00:01:15,236 Which makes complete sense for responsiveness. 22 00:01:15,976 --> 00:01:18,976 But, uh, the actual packaging for the release... 23 00:01:20,076 --> 00:01:22,256 something went sideways. 24 00:01:22,266 --> 00:01:27,456 They- they literally forgot to package the codex-code-mode-host binary in the macOS 25 00:01:27,516 --> 00:01:29,006 and standalone installers? 26 00:01:30,196 --> 00:01:32,136 Yeah, a total packaging omission. 27 00:01:32,736 --> 00:01:38,186 The build scripts ran, the main TUI wrapper looked for its sibling binary at the 28 00:01:38,296 --> 00:01:42,776 designated path, found absolutely nothing, and just panicked. 29 00:01:43,416 --> 00:01:48,906 It was a classic "it worked on my machine" scenario because the dev environments had 30 00:01:48,936 --> 00:01:51,456 the local binary symlinked already. 31 00:01:52,114 --> 00:01:56,554 Okay, so enter v0.144.1. 32 00:01:57,194 --> 00:01:59,754 I saw they patched this almost immediately. 33 00:02:00,354 --> 00:02:03,814 How did they actually secure the launch chain this time, besides just, 34 00:02:03,854 --> 00:02:05,844 you know, remembering to include the file? 35 00:02:06,551 --> 00:02:07,461 Well, two things. 36 00:02:08,091 --> 00:02:13,131 First, they fixed the macOS symlinking in the installer package so the pathing 37 00:02:13,231 --> 00:02:14,631 actually resolves. 38 00:02:14,651 --> 00:02:18,331 But the real engineering fix is the new silent fallback. 39 00:02:18,971 --> 00:02:25,271 If the TUI attempts to bootstrap and realizes the codex-code-mode-host binary is 40 00:02:25,331 --> 00:02:28,451 missing or corrupt, it doesn't crash anymore. 41 00:02:28,511 --> 00:02:32,431 It gracefully falls back to an embedded in-process runtime. 42 00:02:32,450 --> 00:02:32,939 Huh. 43 00:02:33,739 --> 00:02:38,359 If they can just run it in-process anyway as a fallback, why go through the trouble 44 00:02:38,399 --> 00:02:40,899 of the separate sibling process at all? 45 00:02:40,939 --> 00:02:43,179 Is the performance hit that bad on the fallback? 46 00:02:43,953 --> 00:02:47,992 It's- it's- it's noticeable if you're doing heavy multi-file refactoring. 47 00:02:48,653 --> 00:02:52,693 The UI frame rate drops because you're sharing that single-threaded runtime loop 48 00:02:52,733 --> 00:02:53,013 again. 49 00:02:53,593 --> 00:02:56,253 It's meant as a safety net, not the target state. 50 00:02:56,833 --> 00:03:02,033 And speaking of state, they also upgraded the codex doctor command to help diagnose 51 00:03:02,073 --> 00:03:03,853 this exact kind of environment mess. 52 00:03:04,598 --> 00:03:06,238 Oh, the doctor tool. 53 00:03:06,798 --> 00:03:09,158 I used that to debug my node setup. 54 00:03:09,398 --> 00:03:12,338 Did they fix the pnpm detection issue? 55 00:03:12,448 --> 00:03:17,008 It used to complain about missing global packages even when they were clearly there. 56 00:03:18,115 --> 00:03:19,075 Yes! 57 00:03:19,135 --> 00:03:23,215 They corrected the global pnpm package manager resolution logic. 58 00:03:23,715 --> 00:03:29,215 The doctor tool was looking in the wrong global node_modules path for pnpm setups, 59 00:03:29,775 --> 00:03:31,135 flagging false positives. 60 00:03:31,675 --> 00:03:32,905 That's finally resolved. 61 00:03:33,630 --> 00:03:34,180 Nice. 62 00:03:34,750 --> 00:03:39,360 And there were a couple of other nice QoL updates tucked into 144.0, 63 00:03:39,470 --> 00:03:39,850 right? 64 00:03:39,910 --> 00:03:41,190 Like MCP auth? 65 00:03:42,147 --> 00:03:47,047 Yeah, interactive Model Context Protocol authentication is now graduated. 66 00:03:47,627 --> 00:03:51,867 You don't need those clunky experimental flags anymore to authenticate. 67 00:03:51,907 --> 00:03:57,327 Plus, they added automatic sanitization for pasted terminal sequences-so if you 68 00:03:57,387 --> 00:04:02,507 copy-paste code with hidden control characters, it won't execute raw garbage in your 69 00:04:02,527 --> 00:04:02,827 shell. 70 00:04:03,507 --> 00:04:08,427 And they widened the skill-name TUI layout so your custom tool names don't get 71 00:04:08,487 --> 00:04:08,827 clipped. 72 00:04:09,439 --> 00:04:10,719 That's a lot of polish. 73 00:04:11,399 --> 00:04:17,458 If you're on 144.0, run codex update or check your symlinks manually if you're on a 74 00:04:17,518 --> 00:04:18,898 manual Mac install. 75 00:04:18,958 --> 00:04:19,699 That should get you sorted. 76 00:04:20,279 --> 00:04:20,958 Good chatting, Ethan. 77 00:04:21,884 --> 00:04:23,045 Yep, catch you next time.