Scanning never creates a Roll, so scanned frames lose every roll-scoped feature

[kurtwindisch]

I wanted to explore the fact that scanning a roll via either the film scanner or camera scan interfaces never creates a roll in the library. This first seemed like an inconvenience, adding an extra step of having to refresh the library to discover the roll folder. But with the the new roll-oriented workflow, it seems like there could be larger problems as a result of this gap/bug.

Here is a summary of the analysis so far. I'm still trying to understand the degree of any issues (and whether the impacts below are correct) and would like to know what others think.

---

Summary

Frames from the Film Scanner and from camera scan land in the Film Strip but never
become a Roll. state.active_roll_id stays unset, so the Roll panel, roll defaults,
Half Frame, Roll Analysis and scenes are all unavailable on exactly the frames a
scanning session produces. Now that the workflow starts from the Roll panel, this is
the one entry point that does not reach it.

What creates a Roll today

Three functions, and nothing else: rolls.recognize_folder (rolls.py:71),
rolls.import_subfolders_as_rolls (rolls.py:139) and rolls.create_virtual_roll
(rolls.py:204). Their only callers are open_library_folders (controller.py:1475),
import_subfolders_as_rolls / rediscover_rolls (controller.py:1446,
controller.py:1458) and create_roll_from_session (controller.py:1514).

Both scan paths call request_asset_discovery(...) directly and reach none of them:

• _on_scan_finished / _on_scan_batch_finished — controller.py:4872, controller.py:4878
• _on_capture_finished — controller.py:5290

The two scanners are in different shape:

• Camera scan already has a roll boundary. The Scanlight panel has a Roll field and
  writes to <hot folder>/<roll>/ (scanlight.py:1337). The name reaches
  config.metadata.capture_roll for XMP and the {{ roll }} export token
  (controller.py:2383, xmp.py:101) and stops there. It never reaches the roll store.
• The Film Scanner has no roll boundary at all. It writes flat into one Output
  folder under a {{ date }}_{{ seq }} pattern (settings.py:40, scan.py:1403,
  service.py:134), and one batch is one strip, not a roll. A 36-exposure roll is
  several batches into the same folder.

Problems

1. Half Frame cannot be turned on. half_frame.py:98 disables Enable, Adjust and
Auto without a roll, and request_asset_discovery passes half_frame=False when
active_roll_id is unset (controller.py:1371). A half-frame roll scanned in does not
split. The sticky half_frame_mode flag can be on and discovery still refuses, so the
state is not even visible as "off".

2. Every Roll-tab card reads Frame with the Roll button disabled.
controls_panel.py:948 and metadata.py:422. This covers Film Mode, Calibration, Crop,
Roll Analysis, Metering, Raw Decode, Lens, Flat Field and all Metadata cards. The rig
facts the scan was made with cannot be set once for the batch. Setting frame 1 and using
Apply to Selected is a different gesture: it records no roll default, so Reset to Roll
has nothing to return to.

3. Roll Analysis records nothing. _finish_normalization (controller.py:4262)
pushes the pooled baseline onto the loaded frames, but skips set_roll_normalization
and leaves source empty. The baseline holds for the session only. Scenes are
unavailable, since create_scene needs a roll id.

4. Nothing survives a restart. restore_session recovers the roll by looking the
session paths up in the roll store (controller.py:1316, controller.py:1320). Scanned
paths are in no roll, so the session returns rollless and problems 1 to 3 repeat.

5. Gear auto-match never runs. The Roll Settings prefill hangs off
folder_roll_created, emitted only when a folder is recognized for the first time
(library_tree.py:183 → files.py:1186). A scanned folder is never recognized.

6. Library Refresh recovers only one folder shape. rediscover_rolls re-walks only
registered import-source parents (controller.py:1451), and discover_roll_folders
stops descending at the first folder that holds images (rolls.py:123). See the table
below for what that does and does not reach.

7. Scanning while a roll is open absorbs the new frames into it.
_on_discovery_finished appends every newly discovered path to the active roll's
membership (controller.py:2061). Scanning roll 2 while roll 1 is open writes roll 2's
frames into roll 1's extra_paths, with no prompt and no undo. On the Film Scanner this
is not an edge case: the Output folder is sticky, so it happens on the second roll of
every session.

8. Refresh plus an open roll produces double membership. When both 6 and 7 apply,
the scanned frames end up members of two rolls at once: the roll that was open absorbed
them into extra_paths, and Refresh also recognized their folder as its own roll.
rolls_containing_path then returns both, _roll_id_for_restored_paths returns None
because the paths no longer agree on one roll, and the session comes back rollless after
a restart anyway — with one roll's membership permanently polluted.

What Library Refresh does and does not recover

Measured with the scan folder under a parent registered by Import Subfolders as Rolls:

Scan folder	Refresh result
A new sibling subfolder under the parent	Found. Becomes a roll named parent/folder.
A subfolder inside an existing roll folder	Never found — the walk prunes at the first folder holding images.
A new sibling, scanned while a roll was open	Found, and now the frames belong to two rolls (problem 8).

Two further conditions on the first row: the folder must not have been deleted as a roll
before (skip_dismissed=True keeps dismissed folders out), and its name must not match a
discovery filter (default export).

In every case Refresh does not open the roll. active_roll_id stays None, so Half
Frame still did not split the scan that produced the frames, and a Roll Analysis run
before the refresh recorded nothing.

The only full recovery is Import Folder as a Roll on the scan folder, followed by
opening it. That restores everything, and it survives a restart. It is usable where the
output folder is already per-roll (camera scan) and not where it is one sticky flat
folder (the Film Scanner), because importing that folder gives one roll holding every
roll ever scanned into it.

Related finding: "roll" means two different things

The Film Strip's contents and active_roll_id are independent state, and the
roll-scoped operations do not agree on which one they mean:

• Scope on the visible strip, and need no roll id. Reset Roll
  (request_reset_roll, controller.py:4303) acts on
  visible_actual_indices_ordered(). Roll Analysis reads state.uploaded_files
  (_normalization_targets, controller.py:4050) and runs with no roll; only the
  persist step is skipped. Save as Roll takes uploaded_files wholesale.
• Require the roll id. Half Frame, the Frame/Roll scope buttons, Reset to Roll,
  scenes, the stored baseline, roll-scoped edit forks, section_push.
• Resolve membership per frame, ignoring both. _roll_id_for_orphan_asset
  (session.py:1149) gives a frame its own roll's defaults from whatever roll contains
  that path. A strip holding frames from three rolls gives each frame a different
  roll's defaults, so the model does not assume the strip is one roll.

The two also diverge with a roll open: open_library_folders(add_to_session=True) adds
frames without touching active_roll_id, and a Film Strip filter can hide roll frames
while Reset Roll then acts only on what is visible.

This means the batch-shaped operations already work on a scanned strip; the
identity-shaped ones do not, because they need a roll that persists past the session.

Reproducing it

Use a fresh empty output folder, for example ~/scans/repro/roll_a. Every roll created
below is removable with right-click, Delete, which forgets the roll record only and
leaves files and edits alone.

A. The gap. Film Scanner panel, Output folder ~/scans/repro/roll_a, load a strip,
Scan. Then:

1. The Film Strip tally reads No roll — 6 frames, tooltip "These frames are not one
   roll."
2. Library has no new roll.
3. Half Frame: Enable greyed, hint "Half Frame is a roll-wide setting, and this isn't
   one roll. Open the roll itself to split its scans." A half-frame roll does not split.
4. Any Roll-tab card header: the Roll half of the scope pair is disabled, tooltip
   "These frames are not one roll, so there is no roll to hold a shared value."
5. Roll panel: nothing selected in the picker, Scenes hidden.
6. Run Roll Analysis, then restart: the baseline is gone and the tally still reads
   No roll.

B. Refresh cannot reach it. Click Refresh in Library. roll_a never appears,
because a scan registers no import source.

C. The flat folder merges two rolls (problem 7). Import Folder as a Roll on
roll_a; everything in A now works. Leave it open, load a different roll of film, and
Scan without touching the Output folder. The new frames join the open roll. Library
shows one roll with the combined count, and Refresh will not separate them.

D. Camera scan proves it is only a missing call. Camera Scanning panel, Folder
~/scans/repro/cam, Roll ReproRoll1, scan a few frames. They land in
~/scans/repro/cam/ReproRoll1/ — correctly named, correctly per-roll. The tally still
reads No roll and Library is still empty. Change Roll to ReproRoll2 and scan a few
more. Then Import Subfolders as Rolls on ~/scans/repro/cam: both appear immediately,
correctly named and separated. The roll boundary and its name were on disk all along.

E. Double membership (problem 8). Import Subfolders as Rolls on a parent holding one
roll folder. Open that roll. Scan into a new sibling subfolder under the same parent.
Refresh. The new folder becomes its own roll, and its frames are also in the first
roll's extra_paths. Restart: the session comes back rollless.

Design constraint

The roll has to exist and be active before request_asset_discovery runs. Creating
it after the files land does not fix problem 1, because half-frame splitting is decided
inside discovery — which is also why Refresh, which never opens the roll, cannot stand in
for a fix.

Open questions

• Scope: camera scan alone, where the roll name and per-roll folder already exist, or
  the Film Scanner too?
• If the Film Scanner gets a roll: a Roll name field writing to <output>/<roll>/,
  mirroring camera scan; a virtual roll over the flat output folder; or recognizing the
  output folder itself?
• Problems 7 and 8 are membership bugs that stand on their own. Same change, or separate?

[kurtwindisch]

#1256 mostly solves this issue by adding "Scan as a Roll".
The remaining problem that I see is that the roll is imported into the library at the top level of any hierarchy of nested folders that it may actually belong in.