caoer115/ucc
1
0
Fork 0
Code Issues Pull requests 45 Projects Releases Packages Wiki Activity

install.sh: la_put makes ~/Library/LaunchAgents and ~/Library/Logs before each Mac plist write (v12 red, D d436) #69

Closed
caoer115 wants to merge 0 commits from agent/8751a366-darwin-launchagents into main
pull from: agent/8751a366-darwin-launchagents
merge into: caoer115:main
Conversation 1 Commits - Files changed -
caoer115 commented 2026-10-24 17:44:25 +00:00
Owner
Copy link

v12 product red (D d436, both Macs): install.sh stopped at the ucc.cccd-start.plist write on a home with no ~/Library/LaunchAgents. Card: 13-16-ucc-rearchitecture/tasks/v12-darwin-launchagents.

Cause

a524e3f (legacy crossing) moved the follow tick into follow_timer(), which runs after the daemon section and returns early where ucc-follow is the host's own (G1). Its mkdir -p "$LA" "$HOME/Library/Logs" was the only one, so the daemon's start plist (:2459 at 3c5b53f) was written before any folder existed. In v11 (39737ee) the mkdir ran first.

Fix

la_put (install.sh:2354) makes ~/Library/LaunchAgents and ~/Library/Logs (where launchd opens the agent's log), then writes the plist. Both write sites use it: ucc.follow (:2384, any floor, where ucc wrote ucc-follow) and ucc.cccd-start (:2464, floors 3 and 4). The mkdir now comes with each write, so it holds on a first install, a reinstall and a crossing, whatever order the sections run in.

Rows (proved by D on this head; the release worker's proof D d437)

  • launchd-dirs-darwin on mac1: ucc stage --scripts then install.sh --floor 3 --release from the published set. It runs stage.new/install.sh :2354 (la_put) through :2464 (cccd-start), then :2384 (follow tick, called at :2476), then la_load on both plists.
  • launchd-dirs-darwin on mac2: install.sh --floor 3 --from <stage> --round, which runs the same lines from the clone's install.sh.
  • Before the install, either folder that exists is moved aside into ~/round-<rNN>/, so each launch runs on a home with neither. The row is scored right after the install, so it also scores when the install fails. pass = rc 0, both folders exist, and both plists are in LaunchAgents. fail gives the rc, what is missing, and the install's No such file or directory line. n = only when a folder could not be moved aside. (rounds/darwin.sh:171-196, manifest +2.)

Gates at 7162fc1

  • bash -n rc 0 on install.sh, ucc and rounds/darwin.sh, under bash 5.3.15 and 3.2.57.
  • lint-lcall.awk install.sh and lint-lockfd.awk ucc install.sh: clean, no lines.
  • Local check under bash 3.2.57 on a scratch home with no ~/Library. la_put taken from install.sh wrote both plists and made both folders. The row block taken from darwin.sh scored pass on a stub install that makes them, and fail (rc 1 plus the :2459 line) on the d436 output. A Linux host cannot run the darwin path itself; D proves it.

Step 2: darwin writes and launchctl calls read in 39737ee..3c5b53f

  • follow_timer (moved by a524e3f, G1-gated): the break described above. Fixed.
  • The ucc.cccd-start write: same position as in v11, but now the first write into $LA: the break. Fixed.
  • la_load / la_unload / la_still (launchctl bootstrap|bootout|print): bootstrap reads $LA/<label>.plist and runs only after la_put at both sites. Bootout and print need no file.
  • Floor < 3 on a Mac (:2471): bootout, then rm -f behind [ -e ]. No write.
  • XSET (:934): L0 copies the two plists where they exist (- where they don't). The put-back (xput) renames a copy back only into the folder it was taken from, and removes the run's plist for a - row. No write ahead of its folder.
  • Put-back step 3 (:1349-1350): bootout of the jobs this run loaded. No file needed.
  • Darwin reads added in the range: xown's lsof -d txt, the port census's lsof path, and exec -a in the put-back's old-daemon restart (works under bash 3.2.57). All reads; none needs a folder.
  • A different issue, not changed here: after a put-back, the old plists are back as files, but neither is loaded again (la_load is only at the two install sites). On a Mac where the crossing booted out an old ucc.cccd-start or ucc.follow, that job stays unloaded until the next login.

Do not merge from this seat; the leader merges after the Fable correctness and security reviews of install.sh.

v12 product red (D d436, both Macs): `install.sh` stopped at the `ucc.cccd-start.plist` write on a home with no `~/Library/LaunchAgents`. Card: `13-16-ucc-rearchitecture/tasks/v12-darwin-launchagents`. ## Cause `a524e3f` (legacy crossing) moved the follow tick into `follow_timer()`, which runs after the daemon section and returns early where `ucc-follow` is the host's own (G1). Its `mkdir -p "$LA" "$HOME/Library/Logs"` was the only one, so the daemon's start plist (`:2459` at `3c5b53f`) was written before any folder existed. In v11 (`39737ee`) the mkdir ran first. ## Fix `la_put` (`install.sh:2354`) makes `~/Library/LaunchAgents` and `~/Library/Logs` (where launchd opens the agent's log), then writes the plist. Both write sites use it: `ucc.follow` (`:2384`, any floor, where ucc wrote `ucc-follow`) and `ucc.cccd-start` (`:2464`, floors 3 and 4). The mkdir now comes with each write, so it holds on a first install, a reinstall and a crossing, whatever order the sections run in. ## Rows (proved by D on this head; the release worker's proof D d437) - `launchd-dirs-darwin` on **mac1**: `ucc stage --scripts` then `install.sh --floor 3 --release` from the published set. It runs `stage.new/install.sh` `:2354` (`la_put`) through `:2464` (cccd-start), then `:2384` (follow tick, called at `:2476`), then `la_load` on both plists. - `launchd-dirs-darwin` on **mac2**: `install.sh --floor 3 --from <stage> --round`, which runs the same lines from the clone's `install.sh`. - Before the install, either folder that exists is moved aside into `~/round-<rNN>/`, so each launch runs on a home with neither. The row is scored right after the install, so it also scores when the install fails. **pass** = rc 0, both folders exist, and both plists are in LaunchAgents. **fail** gives the rc, what is missing, and the install's `No such file or directory` line. **n** = only when a folder could not be moved aside. (`rounds/darwin.sh:171-196`, manifest +2.) ## Gates at 7162fc1 - `bash -n` rc 0 on `install.sh`, `ucc` and `rounds/darwin.sh`, under bash 5.3.15 and 3.2.57. - `lint-lcall.awk install.sh` and `lint-lockfd.awk ucc install.sh`: clean, no lines. - Local check under bash 3.2.57 on a scratch home with no `~/Library`. `la_put` taken from `install.sh` wrote both plists and made both folders. The row block taken from `darwin.sh` scored pass on a stub install that makes them, and fail (rc 1 plus the `:2459` line) on the d436 output. A Linux host cannot run the darwin path itself; D proves it. ## Step 2: darwin writes and launchctl calls read in `39737ee..3c5b53f` - `follow_timer` (moved by `a524e3f`, G1-gated): the break described above. Fixed. - The `ucc.cccd-start` write: same position as in v11, but now the first write into `$LA`: the break. Fixed. - `la_load` / `la_unload` / `la_still` (`launchctl bootstrap|bootout|print`): bootstrap reads `$LA/<label>.plist` and runs only after `la_put` at both sites. Bootout and print need no file. - Floor < 3 on a Mac (`:2471`): bootout, then `rm -f` behind `[ -e ]`. No write. - XSET (`:934`): L0 copies the two plists where they exist (`-` where they don't). The put-back (`xput`) renames a copy back only into the folder it was taken from, and removes the run's plist for a `-` row. No write ahead of its folder. - Put-back step 3 (`:1349-1350`): bootout of the jobs this run loaded. No file needed. - Darwin reads added in the range: `xown`'s `lsof -d txt`, the port census's `lsof` path, and `exec -a` in the put-back's old-daemon restart (works under bash 3.2.57). All reads; none needs a folder. - A different issue, not changed here: after a put-back, the old plists are back as files, but neither is loaded again (`la_load` is only at the two install sites). On a Mac where the crossing booted out an old `ucc.cccd-start` or `ucc.follow`, that job stays unloaded until the next login. Do not merge from this seat; the leader merges after the Fable correctness and security reviews of `install.sh`.
a524e3f moved the only mkdir of both folders into follow_timer, called after the daemon's start plist and skipped where ucc-follow is the host's own (G1), so on a home with no ~/Library/LaunchAgents (both D users) install.sh stopped at the ucc.cccd-start.plist write. la_put ties the mkdir to each plist write: ucc.cccd-start (floor 3 and 4) and ucc.follow (every floor), on a first install, a reinstall and a crossing.

Rows: launchd-dirs-darwin on mac1 (--release from the published set) and mac2 (from the stage), each on a home with neither folder (one standing is moved aside into the round folder), scored right after the install. CHANGELOG § v12: the row; install.sh 2517 → 2522 lines, the manifest 320 → 322.
caoer115 commented 2026-10-24 18:44:04 +00:00
Author
Owner
Copy link

Merged to main as ee28a52 (pushed --no-ff merge, tree = reviewed head 7162fc1; Fable correctness and security: no findings; proof D d437 0 fails on both Macs; bash -n and both lints clean; zero round scopes on nyc-2 at push). Closing as merged.

Merged to main as ee28a52 (pushed --no-ff merge, tree = reviewed head 7162fc1; Fable correctness and security: no findings; proof D d437 0 fails on both Macs; bash -n and both lints clean; zero round scopes on nyc-2 at push). Closing as merged.
caoer115 closed this pull request 2026-10-24 18:44:04 +00:00

Pull request closed

Please reopen this pull request to perform a merge.
Sign in to join this conversation.
Reviewers
No reviewers
Labels
Clear labels
No items
No labels
Milestone
Clear milestone
No items
No milestone
Projects
Clear projects
No items
No project
Assignees
Clear assignees
No assignees
1 participant
Notifications
Due date
The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set.

Reference
caoer115/ucc!69
No description provided.