guidePrograms

Channels per program request. 25 channels x 24 h is ~460 programs on the fixture server.

Rows kept near either side of the visible rows: one whole block, so the next block is already loading by the time the user reaches it.

Blocks that may hold programs at once. At most four are ever wanted (the visible rows plus BUFFER_ROWS either side can touch four blocks), so this keeps a couple of recently visited blocks for scrolling back without refetching.

3 h

A block is fetched again only once it covers less than NEED_BEFORE / NEED_AFTER around the left edge, so pressing Left or Right does not start a fetch each time.

Programs more than this outside the window are trimmed.

6 h

The time window blocks load programs for: WINDOW_BEFORE before the left edge to WINDOW_AFTER after it (seconds).

The block a row belongs to.

The rows of block present in the grid, as [first, last] inclusive.

Blocks to drop so at most cap hold programs: never a wanted one, farthest from the focused block first. The guide passes CAP.

The fetch that brings a block holding state up to the window: only the side it is missing when what it holds overlaps the window, else the whole window in place of what it holds.

Whether every row of block has arrived.

Whether key's block covers enough of the window to need no fetch. A stale block never does.

Whether block is one of wanted.

The next fetch to make, or invalid when every wanted block covers the window or is waiting to retry. Fetches are made one at a time; call again after each lands.

When the earliest wanted block waiting to retry may be fetched, or invalid when none is waiting: the time to come back if nextFetch has nothing to do now.

failures with only the wanted blocks' entries, so a block the focus comes back to later starts again from the first, shortest delay.

What rows in [firstRow, lastRow] without programs should say. TimeGrid has one text for every such row. FAILED wins over LOADING: a block waiting up to 30 s to retry must not read as loading, while a block still loading next to it reads as unavailable only until it lands.

The time window blocks should hold programs for, from the grid's left edge.

need no fetch; keepFrom / keepUntil, outside which what a block holds is trimmed

What to cut back so blocks hold programs only near the window, farthest from the focus first: a block no longer wanted that holds anything outside [keepFrom, keepUntil] is dropped whole (coming back to it loads its own window anyway), as is a wanted one that no longer overlaps the window at all; any other wanted one is cut back to the window on the side that ran past it. The drops come before the cuts.

until to keep

Blocks to hold programs for, nearest the focused row first: the focused block, then alternately below and above it. Only blocks whose rows have all arrived are listed: a block fetched while its channels are still paging in would miss the late ones.