retire the per-registry control-session name once a psmux release carries the psmux#599 fix #733

Activity

1. pbean

   added

   blocked:upstreamWaiting on a change in another repositoryWaiting on a change in another repository

   area:psmuxpsmux terminal-multiplexer backendpsmux terminal-multiplexer backend

   P4Parked - needs design, blocked upstream/PR, or speculativeParked - needs design, blocked upstream/PR, or speculative

   on Sep 20, 2026
2. pbean

   added a commit that references this issue on Oct 6, 2026

   test(psmux): tripwire probes for three upstream fixes awaiting release

   ...

   076937a
3. dracic

   added 2 commits that reference this issue on Oct 13, 2026

   refactor(runs): return the psmux control session to one fixed name

   ...

   This issue will close once commit 9159f00 is merged into the 'main' branch.

   9159f00

   refactor(runs): return the psmux control session to one fixed name

   ...

   This issue will close once commit 28a9eda is merged into the 'main' branch.

   28a9eda
4. pbean

   closed this as completedin 28a9edaon Oct 13, 2026
5. dracic commented on Oct 13, 2026

   

   dracic

   on Oct 13, 2026

   ContributorAuthor

   More actions

   cleanup and attach never mention a pre-upgrade bmad-loop-ctl-<digest> control session after the rename (psmux 3.4.0)

   After the rename in #900, a control session started under the old bmad-loop-ctl-<16 hex> name keeps running in the project's own registry, as the CHANGELOG says. What the CHANGELOG does not say is that cleanup does not report this surviving session in the primary registry, and attach does not locate its window. With such a session holding a parked run window tagged for the project, bmad-loop cleanup --dry-run prints nothing to clean up, cleanup --json reports "legacy_leftovers": [], and bmad-loop attach <that run> answers nothing to attach (rc 1). The legacy session and its run window remain on that session's psmux server. The docs tell the operator to "end the old session once its windows have parked", but an operator who did not read the CHANGELOG is told, by these cleanup and attach commands, that there is nothing there.

   Measured on psmux 3.4.0 (20aa1d7), Windows 11 10.0.26200, bmad-loop 4c00c37, three runs plus a positive control. The block below is launched from PowerShell 7 (SHELL removed) in a bmad-loop checkout with uv sync done; zero LLM tokens. The legacy session is seeded through bmad-loop's own PsmuxMultiplexer, the way the old release left it: the old digest name, plus a parked run window tagged for the project with set_window_option(wid, runs.PROJECT_OPTION, runs.project_tag(project)). The positive control seeds the identical window under the current name.

   Remove-Item Env:SHELL, Env:PSMUX_DATA_DIR -ErrorAction SilentlyContinue
   $env:BMAD_LOOP_MUX_BACKEND = 'psmux'; $env:PSMUX_NO_WARM = '1'; $env:PYTHONUTF8 = '1'
   $env:BMAD_LOOP_STATE_DIR = "$env:TEMP\bl-legacy\state"
   $Proj = "$env:TEMP\bl-legacy\proj"
   New-Item -ItemType Directory -Force $Proj | Out-Null; git -C $Proj init -q -b main
   Set-Content "$Proj\src.txt" x; git -C $Proj -c user.email=t@t -c user.name=t add -A; git -C $Proj -c user.email=t@t -c user.name=t commit -q -m init
   $rid = '20261013-190001-abcd'
   New-Item -ItemType Directory -Force "$Proj\.bmad-loop\runs\$rid" | Out-Null; Set-Content "$Proj\.bmad-loop\runs\$rid\state.json" '{}'
   
   # Seeds a control session in the project's own registry, through bmad-loop's own adapter:
   # a parked run window tagged for the project exactly as bmad-loop tags it.
   # "legacy" = the bmad-loop-ctl-<digest> name the previous release minted; "current" = bmad-loop-ctl.
   $seed = @'
   import hashlib, sys
   from pathlib import Path
   from bmad_loop import runs
   from bmad_loop.adapters.multiplexer import get_multiplexer
   project, rid, which = Path(sys.argv[1]), sys.argv[2], sys.argv[3]
   root = runs.export_psmux_registry_root(project)  # the project's derived registry
   name = "bmad-loop-ctl-" + hashlib.sha256(root.lower().encode()).hexdigest()[:16] if which == "legacy" else "bmad-loop-ctl"
   mux = get_multiplexer()
   mux.new_session(name, project)
   wid = mux.new_parked_window(name, "run-" + rid, project, ["pwsh", "-NoProfile", "-Command", "exit 0"], "@bmad_return")
   mux.set_window_option(wid, runs.PROJECT_OPTION, runs.project_tag(project))
   print(root)
   print(name)
   '@
   
   $Reg, $Legacy = uv run python -c $seed $Proj $rid legacy; Start-Sleep 3
   uv run bmad-loop cleanup --dry-run --project $Proj    # rc 0: nothing to clean up
   uv run bmad-loop cleanup --json --project $Proj       # rc 0: a real cleanup; every list empty, "legacy_leftovers": []
   uv run bmad-loop attach $rid --project $Proj          # rc 1: nothing to attach for run 20261013-190001-abcd
   $env:PSMUX_DATA_DIR = $Reg; psmux list-windows -t "=$Legacy" -F '#{window_id} #{window_name}'
   #   @1 shell
   #   @2 run-20261013-190001-abcd                        <- still there after the real cleanup
   psmux kill-server; Remove-Item Env:PSMUX_DATA_DIR
   
   # positive control: the same window, seeded the same way, under the name bmad-loop addresses now
   $null = uv run python -c $seed $Proj $rid current; Start-Sleep 3
   uv run bmad-loop cleanup --dry-run --project $Proj    # rc 0: would close ctl window run-20261013-190001-abcd
   $env:PSMUX_DATA_DIR = $Reg; psmux kill-server; Remove-Item Env:PSMUX_DATA_DIR

   	actual	expected
   cleanup / cleanup --dry-run	nothing to clean up, legacy_leftovers: []	the legacy control session named, as cleanup already names a surviving control session in a legacy registry (left in <registry> (not migrated): …)
   attach <run whose window is in it>	rc 1, nothing to attach	either attaches, or says the run's window is in the pre-upgrade session and names it
   the window and session	still running after a real cleanup	still running is fine (CHANGELOG); being unreported is the gap

   Observed at symbol level as well, 3/3: runs.legacy_registry_leftovers(project) returns ({}, []), launch.prunable_ctl_windows ([], []), launch.ctl_window_lookup (None, 0) for the legacy window. At symbol level, kill_ctl_window returns 0 while the legacy window remains; a TUI consequence is source-derived, not independently observed here. Under bmad-loop-ctl the same window gives ctl_window_lookup ('bmad-loop-ctl:@2', 0) and prunable_ctl_windows (['run-20261013-190001-abcd'], []), so the outcome flips on the session name alone.

   Source, bmad-loop 4c00c37. runs.legacy_registry_leftovers (src/bmad_loop/runs.py:2309) is the one reader that names a surviving control session to the operator, but it only walks _legacy_registries() (runs.py:2478: get_multiplexer().legacy_registries()), that is, psmux's default registry and a displaced root:

   # src/bmad_loop/runs.py:2426, :2441
           legacies = _legacy_registries()
   ...
               if is_ctl_session_name(legacy.session_name_key(name)):

   The digest-named session lives in the project's own derived registry, the primary one, which no leftover reader lists. cmd_cleanup (src/bmad_loop/cli.py:6706) prints nothing to clean up from those empty lists:

   # src/bmad_loop/cli.py:6802-6804
       if args.dry_run:
           if not killed and not windows:
               print("nothing to clean up")

   attach_plan calls ctl_window_lookup (src/bmad_loop/tui/launch.py:467), which lists ctl_session(project) only, and that is bmad-loop-ctl since ctl_session_for (runs.py:1805) returns CTL_SESSION (runs.py:1824):

   # src/bmad_loop/tui/launch.py:494-496
       rows = _list_ctl_windows(
           get_multiplexer(), ctl_session(project), ["window_id", "window_name", runs.PROJECT_OPTION]
       )

   is_ctl_session_name (runs.py:1827) still recognises the digest shape, so the prune spares it. Sparing it is correct; nothing reports it.

   Suggested, not validated: in legacy_registry_leftovers, or in a second reader that cmd_cleanup and the TUI's cleanup toast already print beside it, also list ctl-shaped sessions in the primary registry other than ctl_session_for(project). That puts the remainder in the same left in <registry> (not migrated) line and the same legacy_leftovers JSON field. cmd_attach could say "this run's window may be in " when it finds nothing and such a session exists.
6. dracic

   added 4 commits that reference this issue on Oct 14, 2026

   fix(cleanup): report a pre-upgrade bmad-loop-ctl-<digest> control ses…

   ...

   b18bc67

   fix(cleanup): find a pre-upgrade control session through an alias of …

   ...

   5ddb8e8

   fix(cleanup): find a pre-upgrade control session in an honoured regis…

   ...

   170c36e

   fix(cleanup): print a pre-upgrade control session's command only on p…

   ...

   bed31ee