[GLinnik21]

This is the vertical counterpart to the horizontal equal-spacing mechanism (plx_ui::hstack, merged in #886, #888–#891 and #896). It moves the hand-summed y ladders (y += h + GAP) onto one shared stack. No pixel moves.

Mechanism

• plx_ui::vstack (new, ui/src/vstack.rs): a thin facade over hstack's engine, with no second copy of the arithmetic.
  • VStack provides new, heights, height, placed and span.
  • VFlow places lines one at a time as they are painted.
  • Anchor is Top, Bottom or Centre, and VSpan is the placed result.
• hstack::Item::before(gap): an item can set the gap in front of it. The gap between two items is b.before, else a.after, else the stack's own gap. Flow::place debug-asserts that no before is set.
• vstack::AlertHead (with ::ruled): the eyebrow, title and path ladder shared by the tracks panel, the person biography and the About panel. Each panel now measures (head_h / body_rect) and paints from the same stack instead of keeping two copies.
• landing_hero::stack_top is removed. Its callers use hero_stack.

Sites moved onto VStack / VFlow

• ui:
  • StatusOverlay bands and note band
  • still_line
  • the Watchlist read-out
  • hero_chain (detail layout)
  • the landing hero
  • the decision alert
  • route_screen's narrative flow
  • the table_screen hint
  • card_row tile labels
  • collection_tile
  • document_reader
  • the login QR
  • the table two-line row
• appkit: info_panel, up_next, chapters_panel.
• screens:
  • the tracks, person bio and About panels
  • the person header
  • the catalog hero
  • the people lines
  • the About columns
  • the episode meta
  • the login code and status
  • the PIN pad
  • the search empty pair
• app: the diagnostics header, sections and fields ladder.

Left as they are

• The Collection header (header_ys). Its meta and summary offsets are absolute cap tops. Expressed as a stack, the gap would depend on the font, so it would not be a token.
• The filmography preview top. It is a single item, so it stays a plain sum.

Evidence

• Gates on the committed tree (git status empty before and after):
  • make check: every test result: line ok, 0 failed. The largest suites are 1846 passed with 64 ignored, then 1669, 1372 and 1120 passed.
  • make test-hostsim: ok, 285 passed.
  • make check-nodef: ok.
  • make check-replay: 6 replays, every one verdict=SAME, diverged=0.
• Simulator plus mock server, base dbe7b014b against head:
  • 46 scenes, diffed in RGB with the clock masked. Each one was matched against several base samples, because animation phase alone moves up to about 90k px between base runs.
  • Byte-identical:
    • Home, the Continue Watching row and shelves
    • every Detail state, including the About sheet and the deep scrolls
    • Library grids
    • a collection and an empty collection
    • Watchlist, with and without items
    • Search, with and without results
    • the profile picker and PIN pad
    • the tracks panels
    • person pages, with and without a bio
    • the decision alert
    • the login QR
    • every Settings sub-page, including two-line rows and the hold hint
    • the legal document reader
  • The animated Collections fan grid never settles, so it is within noise but cannot be proven identical.
• Not covered by a capture:
  • Labelled still tiles (still_line). The mock cannot produce one. A unit test pins the label baseline at its old position (still_line_label_baseline_is_where_it_always_stood).
  • The chapters panel and the diagnostics page. The simulator cannot reach them headless. Both are covered by unit tests.
• One Opus review of the whole change. Its must-fix items (a still_line baseline that moved, and one lane's work missing from the branch) are fixed.
• Doc-claim audit: 0 findings.

🤖 Generated with Claude Code