layer.transform's fit / fitWidth / fitHeight ops took the layer's size
from render::source_size, which returns (0, 0) for layers without a
source item — shapes and text — so the ops matched nothing and silently
did nothing, for the shortcut, the menu item and the command alike.
When the source size is empty and a fit op runs, the size falls back to
the layer's painted content bounds at the current time, and the anchor
goes to the bounds' centre: content bounds live in layer space, where a
shape is centred on the origin and text sits above its baseline, so
(w/2, h/2) would fill the frame but land off-centre. Reset and the
other ops keep the source size, so their defaults are unchanged.
The regression test asserts what the viewer shows — the painted bounds
mapped to comp space span [0, 0, 640, 360] after fit, and fitWidth
spans the width centred vertically — for a rect shape and a text layer.
Co-Authored-By: GLM-5.3-Flash <noreply@z.ai>
Edit > Duplicate replaces the layer selection with the copies, but an
open in-canvas text edit stayed on the original layer: typing kept
editing a layer that was no longer selected. When the duplicate leaves
the edited layer deselected, end the edit the way #2096's Deselect All
does (text_edit::end_edit), in the command every caller — menu, shortcut,
MCP — dispatches through.
Co-Authored-By: GLM-5.3-Flash <noreply@z.ai>
Every undoable edit keeps the previous project for undo, so the comp it touches
is shared and Arc::make_mut deep-copies it: every layer and every keyframe. An
edit therefore costs time in proportion to the comp's size, and a script that
sets many keyframes in a big comp goes quadratic (a 1,600-layer build with
~100k sampled keys took ~45 minutes).
Comp::layers is now a Layers collection backed by Vec<Arc<Layer>>: copying the
stack copies pointers, and only the layer an edit changes is copied. The API
mirrors Vec<Layer>; serialized form is unchanged (a list of layers).
Comp::layer_mut and the iter_mut().filter/find call sites use find_mut /
iter_mut_where so they copy only the layers they actually change.
- prop.addKey in a 400-layer comp: 1.36 ms -> 0.16 ms per key (flat in comp size)
- the 1,600-layer script build: ~45 min -> 47 s, frames pixel-identical
- regression tests: an edit leaves the other layers shared with undo; layers
round-trip as a plain list
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VkpGcKfWyCudfSsQNiWAaG
Resolve the Interpret Footage form against main's split into
interpret_footage_form and project_settings_form, keeping the Cineon
Settings fields and the pointUnits readback. Restore the Arc, rayon
prelude and cineon_frame imports that main's explicit use lists dropped.
Services::full_path now returns None when the working directory can't be
read, instead of the relative path. file.runScript still runs the script
but does not add it to the list, so a later working directory can't make
the entry open a different file.
File > Scripts lists a bundled script even when an installed ScriptUI
panel has the same name, but running it by name opened the panel. A
script now runs before a panel of the same name, and it enters Recent
Script Files.
- Default disk cache size 20 GB (was 100 GB); settings version 3 migrates the stored old
default and keeps a size the user chose.
- The cache keeps max(2 GB, 5% of the disk) free: sysinfo::disk_space (df -Pk / PowerShell
DriveInfo) is read in the background from memory_tick and caps the limit
(DiskCache::set_disk_space); cache.diskStats reports limitBytes.
- Eviction goes down to 90% of the limit, so the LRU sort runs once per tenth of the limit
written instead of on every write at the limit.
- Failed writes are counted (cache.diskStats failed) and the first is logged.
Also fixes#2063.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
effect.cryptomatteAt rendered the effect's input on every call to learn its pixel grid (up
to ~1 s per hover on a 3D precomp). The aux channels and that grid are now kept in a
CryptoProbe on the Session, keyed by the project revision, layer, effect, frame, scale and
render prefs (nested switches, draft shadows). On footage with nothing animated before the
Cryptomatte effect the frame is the source frame, so a still or a held frame keeps one
probe for every comp time. The effect's own Selection doesn't count: the probe never reads
it, so an animated Selection keeps the probe too.
Session::crypto_probe_ready says whether a query answers from the kept probe and changes
nothing. Session::crypto_probe_job builds it on another thread for the viewer; asking for a
job is starting it. Each started job has a generation: it stores its result only while it
is the newest, a newer job or an edit cancels it, and a panic in the build becomes the
probe's error. A cancelled build checks its generation between its stages (reading the IDs,
rendering the effect's input), so it stops early and lets go of what it read. `probeOnly`
makes cryptomatteAt / pickCryptomatte answer {pending} instead of building on the caller.
A pick re-keys the probe (it only edits the Selection), and undo, redo and history jumps
over picks keep it too (each pick is labelled "Pick Cryptomatte"). Any other edit, footage
changes, Open / New and Session::drop_crypto_probe (the eyedropper disarmed) let it go. The
slot's type is crate-private (the Session field stays public, doc(hidden), so other crates
can still build a Session); crypto_probe_stats and crypto_probe_jobs (jobs run, cancelled
ones included) are doc(hidden), for tests.
Found while testing crates/engine/src/cmd/align_data.rs:75 (parse_stamp)
Co-Authored-By: Claude Haiku 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Two names that differ only in case are one file on Windows and on macOS as they
come, and Win32 drops a trailing dot or space, so `clip.png` with `CLIP.png`,
and `a.png` with `a.png.`, still landed on each other: ten entries extracted to
nine files on APFS, one footage item holding the other's picture. Collisions are
now looked up by a key with the case folded and those characters off the end,
while the name written stays as the entry asked.
Found by CodeRabbit on the pull request.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The fallback name was never checked against the names already taken, so an
entry called exactly what the fallback produces took the same file: `media/1/a:b.png`,
`media/2/a_b.png` and `media/3/2-a_b.png` extracted to five files for six
entries, the third overwriting the second, and both items' footage pointing at
it. Deduplication now runs on the name actually written and counts up until it
is free.
The collision predates making the names safe (two entries named `a_b.png` and
`2-a_b.png` collide on their own), but sanitising gives more ways in, so it
belongs with it.
Found by CodeRabbit on the pull request.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Green, Alt (red) over part of the object, then green there again, as the
Layer panel sends them through roto.stroke: the render follows the last
stroke, and Undo / Redo step between the two results.
EffectCraft writes its own templates as `media/<item id>/<file>` entries, and
extraction trusted both halves. `base_name` splits on `/` and `\`, but the item
id was taken with `split('/')` alone, and it is pasted in front of a base name
another item already used. So a second `media/..\..\..\Startup\x/clip.mp4`
entry became `<dir>/..\..\..\Startup\x-clip.mp4`, which on Windows is three
folders up, with the attacker choosing how the name ends.
The id is now used only when it is one, and the name goes through
`safe_file_name`: separators, drive colons and NUL become `_`, and `.` / `..`
become `footage`. Reached from `templates.create` with `path`, from File ▸ New
from Template, and from a template dropped in the user Templates folder.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The CLI installed no logger, so GPU warnings (a declined frame, out of
memory, fallback to the CPU) were dropped and an export silently rendered
on the CPU. Install the engine logger with a `warn` default (RUST_LOG
still overrides). Split out of #1415.
Upstream split lib.rs files into modules and moved engine commands/ to cmd/; this branch's change merges without textual conflicts.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Opening the editor keeps the keyframe and mask-vertex selection as well as the property one,
and Duplicate cleared only the properties. With the copy selected, Clear then deleted a key or
vertex of the original instead of the copy. During an edit the copies now take the selection the
way a click on them does: layer.select's rule, moved into one helper both use. With no edit
open, Duplicate's selection is unchanged.
The engine's text-edit tests repeated one layer-count closure in five places; they now use
the layer_count helper the 3D tests already had. The doc of forwardable_during_text_edit and
of the Type-tool test helper still spoke of Deselect All alone, though the editor now forwards
Duplicate too.
Duplicate passes over the property selection an open edit kept and copies the edited layer,
but left that selection in place on the original. With the copy selected, the next Duplicate
copied the original's effect, shape item or text animator, and Clear would have removed it.
Ending the edit through Duplicate now clears that property selection along with it.
Opening the editor keeps the property selection, so Cmd+D while typing duplicated an
effect, shape item or text animator selected before and left the edit open. An open edit
now comes first, as it does for Clear: Duplicate ends it and copies the edited layer. With
no edit open, selected effects and items keep their precedence, and the command's
description says so.
SusieeTheFemboyandGLM-5.3-Flash
BFlat
Eduardo EspinosaandClaude Opus 5.5
tylerjenningsw
dotCooCoo
arnav-hiwarkarandClaude Opus 5.5
Matej Kozlovský
NetterSilverandClaude Haiku 5.5
Dylan M. Taylor
Anubhav RaiandClaude Opus 5
vangel76andClaude Sonnet 5.5
Christopher Linnett