The grouping landed in the daemon; this is what it looks like. A tile whose shot
was stored as more than one file badges that in the corner — "RAW" when one of
the members is a raw file, since that is the word the person is looking for, and
a plain count otherwise (a live photo, a burst). The lightbox gains a control
that steps through the group's files, which is the only way to reach the RAW now
that the grid shows the shot once.
Deleting says what it will do: the daemon takes the whole group, so the
confirmation names the file count when a selection stands for more files than
tiles.
The gallery could show photos and nothing else: a photo deleted on the phone
stayed on the page for minutes, and a photo you wanted gone had to be deleted
on another device. Both are now handled here.
Deleting needs a uid-addressed request, because `Request::Delete` takes a
mountpoint path and the photos volume is not in the FUSE mount. `TrashNodes`
trashes through the SDK and answers per-uid outcomes, so a partial failure is
reportable; the daemon then drops those rows with `Db::photos_delete`, album
membership included, rather than leaving the gallery to wait for the next
timeline refresh. The GUI removes the tiles immediately with an Undo toast and
puts back anything the server refused — the photos are in Proton's trash and
stay recoverable there either way. Selection is Ctrl/Shift+click, the Select
toggle or Delete in the lightbox.
Remote deletions arrive on their own now. The Drive volume and the photos
volume have separate event streams, so the photos volume gets its own poller
and its own cursor. Trash and delete events remove rows; anything else only
marks the timeline stale, because there is no inode space to converge on this
volume. With the timeline TTL down to 60 s and an explicit Refresh that waits
for the server instead of only clearing the freshness stamp, a photo deleted
elsewhere leaves the grid in about ten seconds.
The Photos grid laid every photo out as a centre-cropped square. That is
why portraits showed black bars and why a panorama was unrecognisable at
tile size: the tile's shape had nothing to do with the photo's.
Rows are now justified. Photos are taken in capture order until they no
longer fit the target height, and the row is then scaled to land exactly
on the right margin, so each tile is its own photo's shape and nothing is
cropped or letterboxed. The last row of a day stays at the target height
rather than being stretched — a day holding two photos should look like
a short day, not like a layout bug. Ratios come from the daemon where it
has one; where it has not, the tile is laid out square and the day
re-flows once a decoded thumbnail proves the real shape, on the existing
debounce so a screenful of decodes costs one re-flow. Extreme ratios are
clamped, so one strip of film cannot decide what a row looks like.
Ctrl+scroll and Ctrl+± now move the target row height, and the grid is
tighter to match: 2 px gaps, smaller heading margins, a 6 px radius and
no card behind each tile, so the page reads as photographs rather than
as a deck of cards.
Thumbnails also stopped disappearing on scroll. The texture cache was
evicting in insertion order — which is precisely what a scroll back up
asks for again — so it is now least-recently-used and holds 1500 rather
than 600, which at the new density is several screens instead of under
two. A batch tops itself up with the photos just past the rows on
screen, in the direction of travel, and once the queue is short the
near-miss tiles are fetched rather than dropped.
relocate() called forget(), dropping the node's by_uid mapping, so
the re-enumeration assigned a fresh inode. The kernel had already
carried the renamed dentry onto the old one and holds it for the
entry TTL, so lookup/getattr/opendir resolved to an inode entries
no longer held and returned ENOENT — while readdir walked children
and reported the new inode. A renamed directory listed in its
parent but could not be entered until the TTL expired.
Rewrite the node in place and invalidate both parents' listings
instead. Also stops the moved node's DB row being deleted and
recreated on every move, which the B1 data-loss fix worked around
nvrl