Every pin move re-rendered the Puppet's input (source, masks and the
effects before it) at full size on the UI thread and deep-copied it
for the mesh overlay: the cache key of a layer's input hashed all of
its effects, so moving a pin of the Puppet changed the key of its own
input. The input key now covers only the effects that run before the
Puppet, and the overlay shares the cached buffer.
Test: input_keys_ignore_later_effects.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_018gnTnH3Jmk5zr4tMq1Mc9f
- Selecting a mask in the Timeline (its row or Mask Path), or a shape's
Path, selects all its points, so dragging one in the viewer moves the
whole mask, as in After Effects. Delete still removes the mask, and
selecting another property deselects the points.
- Mask Feather has the chain link (Constrain Proportions, on by
default, Alt edits one value) and no longer takes negative values;
Scale's link shows as a toggle too.
- Dragging in the Project panel's empty area draws a selection box
(Shift / Cmd adds); a click there deselects.
Tests: selecting_a_mask_selects_its_points,
mask_feather_values_are_linked_and_not_negative,
dragging_in_the_empty_area_box_selects_items.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_018gnTnH3Jmk5zr4tMq1Mc9f
- With a shape layer selected, the toolbar shows its Fill, Stroke and
Stroke Width (with any tool) and editing them changes the layer's
fills and strokes, as in After Effects (new shape.fillStroke, one
undo step per picker or drag). Shapes drawn without a stroke get one
when a stroke colour or width is set, so changing the stroke now works.
- Alt+Shift+P (A, S, R, T) reveals the property it keyed on each layer.
- A click in the time graph below the last layer deselects the layers
(it only deselected keyframes), so a shape tool draws a new shape
layer instead of a mask on the selected image.
Tests: toolbar_fill_and_stroke_paint_selected_shape_layers,
toolbar_stroke_edits_the_selected_shape_layer,
add_position_key_reveals_position,
clicking_below_the_layers_deselects_them.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_018gnTnH3Jmk5zr4tMq1Mc9f
Trim Paths, Merge Paths, Repeater and the other path operations
existed (layer.addShapeItem) but no menu added them to a shape layer.
Like After Effects, a shape layer's Contents row now has "Add:" with
Group, Rectangle, Ellipse, Polystar, Path, Fill, Stroke, Gradient Fill,
Gradient Stroke and the path operations; with a shape group selected
they go into it. layer.addShapeItem gains the empty Path and Gradient
Stroke kinds.
A 3D text or shape layer in a Classic 3D comp now shows Geometry
Options with "Change Renderer…", which opens Composition Settings on
its 3D Renderer tab (Advanced 3D comps show the extrusion options).
Popup menus are as tall as the window allows and open upward near its
bottom: the 20-entry Add menu was cut at 420 px, hiding Trim Paths.
Tests: shape_contents_add_menu_adds_path_operations,
classic_3d_shape_layers_offer_change_renderer.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_018gnTnH3Jmk5zr4tMq1Mc9f
An 11000 x 2200 comp at Full resolution renders on the CPU, and the
frame went to egui as one texture over the device's 8192 px limit:
the upload failed every frame (a panic in debug builds) and the window
stopped drawing. Creating a comp that size at Full did the same.
Frames, snapshots, the channel / exposure view, other viewers, the
Layer and Footage panels and the Roto Brush overlay now go up averaged
down by the smallest whole factor that fits. The full frame stays
readable for pixel readouts.
Tests: frames_over_the_texture_limit_are_averaged_down and
full_resolution_frames_wider_than_the_texture_limit_fit (panicked
before).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_018gnTnH3Jmk5zr4tMq1Mc9f
Since #103, a preview with audio stayed silent until every frame from
the playhead to the end of the preview (or what fits in the RAM
preview) was cached. The prefetch only renders a few frames ahead, so
on a normal work area the sound came near the end or on the second
loop, even when the comp rendered in real time.
The held sound now also starts once the silent preview has kept real
time for a quarter second with the prefetch window ahead cached. A
comp that can't render in real time still waits for the cache, as #103
intended; a stall clears the measured rate so the sound doesn't cut in
and out.
Test: preview_with_audio_sounds_once_rendering_keeps_up (a 30 s comp
sounds within 2 s, without skipping a frame first).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_018gnTnH3Jmk5zr4tMq1Mc9f
Both visualisers read their Audio Layer's samples around the frame
time, but they weren't in TIME_DEPENDENT, so the viewer's layer cache
gave the layer one key for every frame and showed the first frame's
spectrum for the whole preview. Export doesn't use that cache, which is
why it rendered correctly.
Test: audio_spectrum_follows_time_through_the_layer_cache renders a
tone that starts at 0.5 s through a shared layer cache after t = 0 and
compares it with an uncached render (it showed the stale baseline).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_018gnTnH3Jmk5zr4tMq1Mc9f
Clicking an item's Output To file name opened an inline editor on the
name template (`[compName].[fileExtension]`), so there was no way to
pick a folder from there. Choose File… in the menu beside it did open
a dialog, but the project save dialog: filtered to .ecproj and opening
wherever the OS chose.
As After Effects' Output Movie To, clicking the file name now opens a
save dialog at the current output (its folder and name) for the output
format's extension, and the chosen file becomes the output. Choose
File… opens the same dialog. Alt-click still edits the template in
place, and hosts without a save dialog (the web) keep the inline
editor. The desktop `pick_save_file` hook opens in the folder of a
default path.
Test: clicking_the_output_file_name_opens_the_save_dialog.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Save as New Workspace stored the layout and `window.workspace` could
switch to it, but the Workspace submenu was a fixed list of the
built-in workspaces, so a saved one never appeared and could not be
chosen from the menu.
The submenu now has a `savedWorkspaces` dynamic node after the
built-ins (DynCtx carries the saved names), in the in-window and native
menus, with a check mark on the current one as the built-ins have.
The in-window submenu was also too short to show it. egui lays out a
new submenu in a default 400 pt area, and the scroll area inside took
that height and kept it, so Window ▸ Workspace (514 pt) opened with its
last rows (Undocked Panels and below, where saved workspaces go)
scrolled out of view. Submenus now open at their full height and scroll
only when taller than the window.
Tests: saved_workspace_is_listed_in_the_workspace_menu (picks it from
the menu bar), saved_workspaces_are_in_the_native_workspace_menu.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
apply_output checked `channels: rgba` against the module before it
applied `webmCodec`, and choosing AV1 then quietly dropped the channels
to RGB. So `render --format webm --webm-codec av1 --channels rgba` (one
renderQueue.add) succeeded and wrote an opaque AV1 stream.
The codec options now apply before Channels, so the existing "has no
alpha channel" error (how an unsupported channels request is already
reported) refuses the combination; for WebM it names the codec: "AV1
WebM has no alpha channel (VP9 WebM has)". Choosing AV1 alone on an
RGBA VP9 item still switches it to RGB, as changing the format does.
Test: av1_webm_refuses_requested_alpha.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
apply_output switched the format whenever the Output To extension mapped
to a different format, and `.mp4` maps to H.264 only, so
renderQueue.setOutput with an .mp4 path (even the item's own default
name) turned an HEVC or AV1 module into H.264 and the render wrote an
H.264 stream.
The extension now changes the format only when the current format
doesn't write it (OutputFormat::extension): HEVC, AV1 and H.264 all write
.mp4, so they stay; `.mov` on H.264 still picks ProRes and `.mp4` on
ProRes still picks H.264.
Test: output_to_keeps_a_format_that_writes_the_extension.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Their params docs wrote the renderer as `renderer? classic3D|advanced3D`.
Without a `:`, the doc parser reads `|` as "or another key", so
`advanced3D` became an accepted key in the schemas and the unknown-key
check: `comp.new {"advanced3D": true}` succeeded and made a Classic 3D
comp.
The docs now use the value-list form the other commands use,
`renderer?: classic3D|advanced3D`. The key is removed rather than
implemented: it would only duplicate `renderer` (and could contradict
it). A scan of every command's schema found no other key that came from
a value list.
Test: renderer_values_are_not_parameter_keys (not in the schema, rejected
as unknown, `renderer` still works on both commands).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
add_effect ran effect.apply and then one prop.set per `values` entry as
separate commands: each became its own undo step, and a bad value (an
unknown parameter id) returned an error but left the effect, and any
values set before it, on the layer, so a corrected retry landed at
effects/#2.
It now runs them as one engine.batch, named like effect.apply's own undo
step ("Apply Gaussian Blur"): one undo removes the effect with its values,
and a failing step rolls the whole call back. The reply is unchanged.
Test: add_effect_is_one_undo_step_and_all_or_nothing.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
prop.setExpression keeps an expression with a syntax error but disables
it, and says why only in its own reply. MCP set_property answers with
prop.get instead, and prop.get reported `evaluated` / `expressionError`
only for enabled expressions, so set_property, get_property and prop.get
all showed the broken expression as if it were installed.
prop.get now reports a disabled expression too: `evaluated` is the
static value that renders, and `expressionError` is its syntax error.
Test: expression_syntax_errors_reach_the_reply (set_property and
get_property carry the SyntaxError; runtime errors and valid expressions
report as before).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
`effectcraft-cli script FILE` passes the file's text to script.run, which
ran a compiled .jsxbin script as JavaScript and failed with "SyntaxError:
unexpected '@'" at 1:1, while file.runScript already says .jsxbin isn't
supported.
script.run now refuses a script named `.jsxbin` or whose text starts with
the `@JSXBIN@` header, with file.runScript's message (one shared
constant). The check is in the command, so the CLI with or without
--bridge, MCP run_script and the Script Console all report it.
Tests: engine_entry_points (by header and by name), script_file_and_eval
(the CLI exits 1 with the message and no SyntaxError).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Cause: a `DockNode::Stack` laid its panels out from fixed heights (Preview
46 pt) with the rest shared, but only split gutters were interactive, so
the right column's panels could be expanded or collapsed and never resized.
Fix: the gap below a stacked panel drags the nearest open panels above and
below it against each other, neither going below 40 pt. The new sizes are
stored as the entries' heights, so they persist in the layout like split
sizes. The last panel without a height keeps sharing the rest, and when no
panel takes the rest, or space runs short, the heights scale so the stack
fills its column. Split and stack gaps share one `gutter` helper.
Automation ids: `dock.gutter.<path>s<i>` for the gap below stacked entry i.
Test: ui_tabs `stacked_panel_gaps_resize_the_panels` (drag, persistence,
collapse/expand, minimum); dock unit tests `stacked_panels_fill_their_column`
and `dragging_a_stack_gap_trades_height_between_its_neighbours`.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Cause: the click-to-edit fields (hot numbers and revolutions, the Timeline
time display and layer comments, Project rename/comment, the Render Queue
output path, Essential Graphics control names) opened a text edit with
egui's default caret at the end of the text, so typing appended to the old
value. The Timeline time display, layer comment and Essential Graphics name
editors also re-requested focus before checking `lost_focus()`, which hid
the loss, so Enter, Escape and clicking away never closed them.
Fix: a shared `widgets::keep_focus` keeps an open inline editor focused and
selects its whole text on the frame it opens (`widgets::select_all`, which
the layer rename now reuses). Those editors check for a lost focus first,
and Escape now cancels in the time display and the Essential Graphics
name editor.
Test: ui_angles `typing_replaces_the_clicked_value` (degrees and revolutions
in Effect Controls, Timeline and Properties, typed without Ctrl+A);
ui_inline_edit `project_rename_replaces_the_name` and
`timecode_field_replaces_commits_and_cancels`.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Conforming a footage file's audio (Settings ▸ Disk ▸ Conformed Audio
Folder, which preview playback reads) timed each 65,536-frame chunk as
`frame index × ticks per second / rate` in i64, which overflows past
36,310,516 frames: 12:36.47 at 48 kHz (25:13 at 24 kHz, hence the
workaround). Chunks after the overflow decoded at a wrapped, negative
time and came out silent, so preview audio stopped at 12:37.76 (the end
of the last chunk that started in time) in every long file.
Chunk times now come from Tick::from_units (exact, 128-bit). The
conformed files' names include a format version, so files written by
earlier builds are ignored and conformed again.
Test: conform_chunks_keep_their_times_on_long_files (every chunk start
of a 2-hour file at 24–96 kHz); it failed before (multiply overflow).
Checked end to end: a 13-minute file's conformed audio has sound at 760 s
and 790 s.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A frame whose last byte happened to match the superframe marker pattern
(0b110xxxxx, Annex B) was taken by decoders for a superframe with an
index at its end: ffmpeg rejected it ("Invalid le value at
frame_sizes[i]"), libvpx reported a corrupt frame, and later frames lost
their reference. In the reported WebM, 3 of 60 frames (10, 22, 24) ended
in 0xc8, 0xc0 and 0xca. The render still reported success.
Such frames now get a zero byte after the last tile; the last tile's
bool-coded data may end in zero padding (9.2.3).
Test: frames_never_end_in_a_superframe_marker encodes sequences until a
frame needs the padding, checks no frame ends in a marker byte, and that
the padded frames decode exactly with FilmCraft's decoder and with
ffmpeg (clean error log). It failed before ("96x64 q40 frame 22 ends in
0xde"). The ffmpeg IVF oracle is shared with the existing inter test.
Checked with the reporter's steps: 60/60 frames, no errors from ffmpeg's
vp9 or libvpx, RGB and RGB + Alpha.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The Project panel (like other panels) runs comp.open while it draws, so the
new comp became the active comp mid-frame. Later in the same frame the
active, locked viewer drew and recorded the active comp as its own. On the
next frame on_open_comp compared that record with the opened comp, found
them equal and skipped the lock, so the locked viewer switched comps.
The active viewer no longer records its comp while an OpenComp event is
pending (it asks for a repaint instead); on_open_comp then sees the comp
the locked viewer really shows and routes the new one to an unlocked or
new viewer.
The viewers test's click helper now releases a frame after the press, as
a hand does: a press in a tab strip focuses the shown panel first, and a
same-frame press and release refocused it after the tab's own click.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The Pan Behind gesture ignored modifiers other than Ctrl (snapping): it
always moved Anchor Point and Position together and never constrained the
move, and the tool's button had no double-click action.
- Alt held as the drag begins moves the Anchor Point alone (Position is
left as it is, so the layer shifts).
- Shift keeps the move to the axis dragged along most; Move and Pan
Behind now share that rule (one_axis).
- Ctrl+double-clicking the Pan Behind tool's button runs
layer.centerAnchor (Center Anchor Point in Layer Content).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The Pan Behind drag passed its layer to the snapper's exclude list, which
dropped all of that layer's targets: its corners, edge midpoints and centre
as well as the anchor being dragged. The layer's box doesn't move in comp
space while the anchor pans (Position compensates), so those points are
valid targets.
The engine's per-layer snap targets are now one function, layer_targets,
which can leave out the anchor point. The gesture records its layer's
targets without the anchor as the drag begins and passes them to the
snapper (snap_with) as extra targets.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Selecting a property group also selects its layer, and edit.clear only
looked for selected keys, mask vertices, puppet pins and effects before
deleting the selected layers, so Delete with a text animator, mask or
shape group selected deleted the whole layer.
edit.clear now removes the selected removable groups (text animators,
masks, shape items, trackers) of the active comp in one undo step before
falling back to layers. It shares the removal with prop.removeGroup.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The per-line tracking compensation in text_geom used only the animators'
Line Anchor, which defaults to 0% (left) when no animator has one, so
tracking on centred or right-aligned text pushed the line to the right
instead of spreading it around its alignment point.
Without a Line Anchor, a line now grows from where its own paragraph's
alignment pins it: 50% for centred, 100% for right-aligned (a justified
paragraph by its last line), read per paragraph rather than from the
document. The Line Anchor property still overrides it.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
build::default_label picked a new layer's label from its type alone
(comps Sandstone, images Lavender, ...), ignoring the label of the Project
item it was made from.
A layer made from an item (footage, comp, solid, 3D model) now copies the
item's label when it is created; every path that adds an item to a comp
(layer.addItem from the menus, Project panel, Timeline and viewer drops,
models, precomps, imports) builds its layer there. Layers keep their own
label when the item's changes later. Layers without an item keep the type
defaults.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Each Project panel row registered its click / drag area after the twirl,
proxy indicator, label swatch and comment cell drawn on it, so the row sat
on top and took their clicks: a folder's arrow selected the folder instead
of opening it, and the swatch never opened the label menu.
The row's area now goes in first and those widgets sit on top of it. A
single click on the arrow twirls the folder; a click on the swatch opens
the label menu (project.setLabel). Clicks, double-clicks and drags on the
rest of the row work as before; a single click on the comment cell still
selects the row, and a double-click there now edits the comment.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Its setup wrote the Timeline's old global reveal field, which no longer
filters on its own since reveals are kept per layer.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The navigator drew handles on both ends but the whole bar was one drag
area that scrolled. Each end is now a handle (timeline.navigator.start /
.end) that moves that side of the visible span while the other stays:
pulled in, the ruler zooms in; stretched to the track's ends, it shows
the whole comp again. The middle still scrolls at the current zoom.
; (timeline.zoomFrameToggle) zooms in to frame level around the current
time and, pressed again, back out to the whole comp. timeline.zoomFit
stays as the plain fit command (no shortcut). Both go through set_zoom.
Tests: navigator_ends_drag_to_zoom and
semicolon_toggles_frame_level_and_the_whole_comp (real pointer drags).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Alt+wheel's zoom took its lower limit from the current zoom rather than
the comp's fit and kept the ruler's start where the pointer had put it,
so on long comps it stopped zoomed out on a slice from the middle of the
comp (0:05:02–0:10:27 of a 10-minute comp). The = / - zoom used a
different limit again.
Both now go through set_zoom: the limit is the zoom that fits the comp
(fit_pps, the same formula the ruler and the zoom slider use), and
reaching it switches the ruler back to fitting the whole comp from 0.
Test: alt_wheel_zooms_out_until_the_whole_comp_shows (real Alt+wheel
events on a 10-minute comp); it failed before (Some(1.135) px/s, start
still offset).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Layer ▸ Time ▸ Enable Time Remapping (and the layer menu's entry) added
the property and its two keyframes but left the layer twirled closed, so
nothing visibly changed. When the command turns remapping on, the
layers that have it now twirl open on Time Remap, through the same
per-layer reveal as RR. Turning it off reveals nothing.
Test: enabling_time_remapping_reveals_time_remap.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
U (and P, S, R, T, A, E, M, F, L) kept one reveal filter for the whole
Timeline and replaced the set of open layers with the selected ones, so
revealing on one layer closed every other layer and dropped their
reveals.
Each layer now has its own filter (TimelineState::layer_reveal; an open
layer without one shows its whole tree). A shortcut applies to the
selected layers (all layers with none selected) through apply_reveal and
leaves the others as they are; `reveal` stays the last shortcut's filter
for toggling and Shift-adding. U's results replace only the target
layers' earlier picks, so U after UU shows the keyframed properties
again. anim.reveal tells the Timeline which layers it looked at, so a
layer with nothing keyframed closes. Twirling a layer by hand, Ctrl+`
and Collapse All clear the filters they cover.
Test: reveal_shortcuts_leave_unselected_layers_alone.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
#154: renderQueue.add (and render.addOutputModule) applied `format`, then
dropped it before applying the output path, so `--format hevc --out
x.mp4` saw the .mp4 extension and switched the module back to H.264
(and ProRes / VP9 for .mov / .webm). An explicit format now wins and a
mismatched extension follows it (x.webm → x.mp4 for AV1), as choosing a
format in the Output Module renames Output To; an extension that agrees
with the format keeps the name as given. The format error lists hevc and
av1 everywhere (one FORMATS string).
#155: a `comp` that names no composition (name or id) failed with "no
active composition". comp_id now reports it as `no composition "Nope"`
(EngineError::NoSuchComp), and comp_for_queue passes that on instead of
collapsing it.
Tests: an_explicit_format_wins_over_the_output_extension and
an_unknown_comp_is_named_in_the_error (engine). Checked with the CLI:
the files are hevc/hvc1 and av1/av01 and decode cleanly with ffmpeg.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
With installed families whose names contain the query (JetBrainsMono
Nerd Font and its variants) the test expected exactly one match and
failed. It now checks that every match contains the query, the count
agrees, the query filters, and the bundled JetBrains Mono is there with
its Regular style.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The Tools table gains history and script_ui, get_layer's depth,
add_keyframe's comp and the bridge screenshot's max_side and path.
agents_doc_tools_table_is_current (effectcraft-automation) fails when a
tool or an argument of its input schema is missing from the table, as
effects_doc_is_current guards docs/effects.md.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
#139's bounded read_buffer polled the device without waiting and slept up
to 10 ms between polls, adding up to 10 ms to every readback (the desktop
viewer reads every GPU frame back): the gpu crate's tests took 227 s
against 119 s before. It now waits with PollType::Wait and the remaining
deadline as its timeout (a timeout still retires the context), then takes
the result, giving another thread that polled the same device up to 1 ms
to deliver its callback. The tests take 71–79 s.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The Composition viewer ignored drags from other panels. Now, as in After
Effects:
- A Project item dropped on the viewer becomes a layer centred where it
lands (above the selected layer, like any new layer); while dragging, the
viewer is outlined and the layer's frame follows the pointer.
- Media Browser files do the same; files dropped from the OS file manager
do too when the platform reports where the pointer is during the drag
(otherwise they import as before).
- An effect from Effects & Presets goes on the layer under the pointer,
which is outlined while dragging.
layer.addItem and layer.newModel take `position` ([x, y] comp px).
file.import takes `addToComp` with `time`, `index` and `position`;
mediaBrowser.import shares it (file::add_to_comp): layered files go in as
their comps, the comp they were dropped in stays open, and items that
can't be layers are reported in `errors`. viewer::layer_at is the hit
test pick() and the drop share.
Tests: ui_drag_drop.rs (project item and effect dropped on the viewer)
and imported_files_go_into_the_comp_where_they_were_dropped (engine).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A Project item (or Media Browser file) dropped in the Timeline always
went above the selected layer, starting where new layers start.
Now, as in After Effects, a line shows where it goes in the stack (above
the first layer row whose middle is below the pointer, else below the
last), and over the time graph a marker with a timecode tag shows its In
point, which follows the pointer (frame-snapped) or, with Shift, the
current time. The layer reorder drag shares the insertion logic.
layer.addItem and layer.newModel take `index` (1-based stack position);
mediaBrowser.import passes `time` and `index` on, one layer under the
other.
Tests: project_items_land_where_they_are_dropped_in_the_timeline
(ui_drag_drop.rs, real pointer drags: outline, time graph, Shift, below
the last layer) and add_item_at_a_stack_index_and_time (engine).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The desktop app shares egui-wgpu's device, which had no uncaptured-error
handler, so wgpu's default one panicked on an out-of-memory error on
every frame thread; the panicking frame stayed "in progress" forever and
never rendered again. The GPU frame budget was a fixed 1 GiB.
- GpuContext::new logs uncaptured errors once its pipelines are built
(errors while building them still panic there, which the app catches
to fall back to CPU compositing).
- Gpu::within_memory runs GPU work in an out-of-memory error scope; on an
error the result is discarded and uploaded layers and pooled textures
are freed.
- Frame workers render each frame within it; out of memory, the frame
renders on the CPU and the RAM preview keeps half as many GPU frames
(down to 64 MiB), since wgpu can't tell how much video memory there is.
- A frame whose render panics renders again on the CPU, or is released,
so the viewer asks for it again.
Tests: a_frame_whose_render_panics_is_released (failed before),
running_out_of_video_memory_halves_the_gpu_frame_budget and
within_memory_returns_the_work_when_there_is_room.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Commit c9d5e733 (M13.4, hevcenc deblocking) committed 207 files under
crates/hevcenc/target/agent-codec-hevc/: an agent's CARGO_TARGET_DIR with
rlibs, rmetas, build-script outputs and .rustc_info.json (~36 MB). They
were tracked because .gitignore only ignored the root `/target`.
- git rm --cached the directory; nothing references it.
- .gitignore: `/target` -> `target/`, so a per-crate or per-agent target
dir is ignored anywhere in the tree. No tracked file lives under a
directory named `target`, so nothing else becomes ignored.
The blobs stay in history; this keeps them out of every new checkout and
stops the next agent run from committing more.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude
Brandon Thomas
BFlat
Dexter
Jonas Neubauer
EthanSuquet