Compiler hygiene (all platforms):
- Silence warnings across the tree: missing override, -Wreorder ctor
init, -Wshadow, -Wsign-compare, missing switch cases, unused
functions/captures, Qt 6.11 deprecations (QMouseEvent/QDropEvent
accessors, qAsConst, Q_FOREACH over non-shared containers,
AA_UseHighDpiPixmaps) and the .bak/ backup tree removal.
- Fix regressions from the cleanup: missing clip decls in
capi/timeline.cpp, plugin.cpp rename fallout, panel setFocus
ambiguity, QGraphicsItem::pos vs event->position(), duplicate
k_push_button case, boolean test variable.
Windows:
- Qt portability: CommandLineParser is_set/add_option, QTimeZone
systemTimeZone (QTimeZone::LocalTime is 6.11-only), k_progress_* enum.
- Linking: stop adding oakengine to OLIVE_LIBRARIES (import lib plus
oakengine-obj caused multiple definitions); add OAKENGINE_STATIC so
internal consumers no longer reference __imp_* stubs.
- oakengine.ver: export olive::Renderer typeinfo so liboakgl.so can be
dlopened (Linux), DynamicRenderer no longer dlcloses backend libraries
(crash in RenderManager's dtor calling into unmapped memory).
- OTIO runtime: copy DLLs next to every binary on Windows instead of
relying on PATH (0xc0000135 in gtest discovery).
- Headless GL: the runner only has GDI OpenGL 1.1, killing every render
worker. Deploy Mesa llvmpipe as opengl32sw.dll (Qt's software-GL
channel) with QT_OPENGL=software, and let QT_OPENGL override the
AA_UseDesktopOpenGL default. ExportTask fails fast after 8 consecutive
undelivered frames instead of segfaulting or grinding forever;
FFmpegEncoder::write_frame tolerates null frames.
- Tests: GetTempPathA+PID temp dirs, GetLongPathNameA for 8.3 names,
forward-slash normalization when comparing project filenames.
Linux:
- Install libshaderc-dev so oakvulkan compiles GLSL (Vulkan tests).
- Accept UNORM floor-or-round (63/64) in the blit ping-pong test.
- Skip MainWindow construction test on the offscreen QPA (cannot paint
QOpenGLWidget).
Also: oak_cli_transcode gets a 300s ctest timeout, worker logs GL
context version and LoadGraph/render_frame stages, and
docs/plans/eliminate-event-bridge-issues.md (English translation).
- resizabletimelinescrollbar instantiates QMetaType over OakEngineMarker
via timeline.h includes; the declaration belongs there, not serializer.h
- execinfo.h doesn't exist on Windows; the OAK_DEBUG_INVALID_INPUT
backtrace diagnostic is now Unix-only
The scripted Q_DECLARE_OPAQUE_POINTER sweep missed the QtCore/qmetatype.h
include (cascade of stdlib errors) and declared a few types in two headers
(redefinition of IsPointerDeclaredOpaque). Include qmetatype.h everywhere
and declare each handle in exactly one header.
- Linux GCC rejects QMetaType instantiation over incomplete types
(static_assert(sizeof(T))); Q_DECLARE_OPAQUE_POINTER every opaque
OakEngine* handle in the public headers so QList<OakEngineTask*> etc
compile
- Windows: gtest discovery ran the fresh exe before the DLLs were next
to it; use DISCOVERY_MODE PRE_TEST on WIN32 and copy the DLLs for
ctest time
- Linux: Qt6 metatype static_assert requires complete types; declare
OakEngineBlock with Q_DECLARE_OPAQUE_POINTER in the public header
(outside extern "C")
- Windows: gtest discovery runs the freshly linked test exe at build
time and it cannot find the DLLs (0xc0000135); copy oakcore/oakengine/
ffmpeg_bridge DLLs next to the test binaries on WIN32
- macOS: Imath header mixing between otio-install's stale bundled Imath
and Homebrew's Imath broke ImathBox.h (V2h undeclared); bump the OTIO
cache to v2 and drop bundled Imath headers so the system one is the
only set in play
- capi: oakengine_node_output_connection_at/_at_ex returned the source
node as the connection destination; the actual destination is
conn.second.node(). Out-edge enumeration was useless, so the node
view could only draw in-edges and randomly lost whichever edges
needed the out-edge path (random per context build order).
Regression test in oakengine_node_test
- project teardown: Project::clear() pre-notifies node removal while
nodes are fully constructed (observers used to crash on
half-destroyed nodes); childEvent suppresses the removal dance while
clearing; Node::disconnect_all/disconnect_edge get a silent mode for
teardown so no invalidation/events touch dying members
(is_being_cleared); ClipBlock marker disconnect guarded against
dead viewer/markers; ProjectCopier and PreviewAutoCacher drop
project references on Project::destroyed instead of disconnecting
dead objects at shutdown
- preview: add oakengine_preview_request_get_audio_sample_count; the
viewer queried sample count by passing nullptr to get_audio_samples
which rejects it, so all playback audio was silently dropped
- app: fix unterminated input-id memcpy in ResolveGroupInput
(nodeparamviewitem, widgetbridge) that corrupted every parameter id
- app: unsubscribe raw C-API event subscriptions in destructors of
NodeParamViewKeyframeControl, NodeParamViewConnectedLabel and
ExportDialog; playhead events used to fire into dead widgets
(crash when dragging the playhead)
- tests: preview request roundtrip (video frame + audio range) and
free-while-active teardown coverage; env-gated OAK_DEBUG_EDGES /
OAK_DEBUG_INVALID_INPUT diagnostics
- docs: investigation notes in docs/zh/
- nodeparamview: add visited set to get_distance_between_nodes, fixing
unbounded recursion (stack overflow) when the node graph has a cycle
- viewerdisplay: give texture_ a consistent owner via assign_texture();
borrowed queue textures are now retained, created ones freed, fixing
a dangling pointer that corrupted the heap and crashed in the GL driver
- playbackcache: resignal_requests() iterates a copy, handlers may
clear_request_range() while iterating (ASan container-overflow)
- preview C API: preview request ticket lambdas captured the request
state raw; after oakengine_preview_request_free the ticket outlived
the request and the finished callback wrote into freed memory
(heap-use-after-free). The finished flag is now a shared_ptr captured
weakly by the callbacks
- playback: oak_playback_frame regains a timestamp (num/den) filled
from olive::Frame; the viewer queue append no longer uses Rational()
for every frame, which made append_timewise drop all but the first
frame and froze the picture during playback
- mainwindow: open_node_in_viewer refuses sequence nodes; sequences
already have the Sequence Viewer, and saved layouts could otherwise
resurrect a redundant floating Viewer bound to the sequence
- app/ no longer includes engine C++ headers nor holds engine C++ types:
engine access goes through the oakengine C ABI plus C++ wrappers
(oakutil/oaknode.h, oakutil/oakvideo.h) and app-local mirror types
(tooltypes, trackreferencehandle, timelinecommonapp, keyframetypes,
subtitleapp, serializedlayoutinfoapp, nodevaluehandle, sliderdisplaytypeapp)
- engine: new C ABI functions for block/track/clip/transition navigation
and predicates, links, caches, waveform/playback, disk folder,
sequence_track_list, node_free, footage_is_valid, block_get_track,
get_brush; loadotio/saveotio ported to the current engine API
- OTIO is now a required dependency: CI and CD build it on every
platform, FindOpenTimelineIO fixed for OTIO 0.16/0.19 (the old deps
include requirement silently disabled OTIO everywhere), runtime
libraries are bundled into packages and copied next to macOS binaries
(oak_copy_otio_runtime)
- fix ProjectViewModel drag&drop mime read/write size mismatch (segfault)
- unify color label naming (k_olive -> "Oak") in the app-side mirror
- docs: OTIO required, FFmpeg minimum corrected to 6.0 (en/zh)
- gtest suite: 1925 passed, 0 failed
R7-A (Qwen 3.8 Max): oakengine/display.h rewritten to the POD contract
from r7-pure-abi-plan.md - oak_video_params everywhere, opaque
texture/frame handles with retain/free protocol (engine-heap control
blocks), OakSharedBuffer refcounted wrapper for the QVariant playback
path. All TexturePtr/FramePtr gone from app (47 sites).
R7-B (Qwen 3.8 Max): liboakengine.so exports 3486 -> 19 C++ symbols
(version script oakengine.ver: oakengine_* plus the documented
oakgl/oakvulkan dlopen plugin ABI). oakengine-obj OBJECT library feeds
both the shared lib and the test binaries (-rdynamic so dlopen'd
backends resolve engine objects).
Fix (Kimi K3): producer/consumer type mismatch - viewerdisplay
unpacks OakSharedBufferPtr but viewer.cpp pushed raw void* handles,
so no frame ever reached the display widget (all 5 vulkan viewer
tests timed out with 'never received a texture'). Producers now wrap
with oak_make_shared_frame / oak_make_shared_texture(retain).
Verified: build 0 errors, ctest 45/45, nm U _ZN5olive = 0 in
oak-editor/oak-render-worker/oak-cli, 19 exported C++ symbols in
liboakengine.so (all documented plugin ABI).
The full pure-C facade used by the app: node/project/timeline/viewer/
undo/task/events/serializer/playback/preview/renderer/gizmo/color/
audio/footage/proxy/encoding/exporter/config/disk/ipc/plugin/worker
families, plus undo-group semantics, display renderer handles,
NodeFactory accessors, and per-family pure-C engine tests.
- new primitives: clip_toggle_enabled (per-block flip), clip_set_linked,
sequence_add_default_transition (config-driven, sequence timebase),
node_set_label_many, node_set_color_label, plus observation getters
- toggle-links, default transitions, enable toggles, color labels, and
block renaming now go through the facade; the stray empty undo entry
from block renaming is gone along the way
- nest/multicam/waveform-sync stay as documented composites: they mix
redo_now intermediate state, graph surgery, and app-side computation
that a single primitive cannot express faithfully
- the timeline panel's command execution paths are now fully migrated
- new primitives: keyframes_set_time_many (conflict-safe batch time
move), keyframes_set_value_many (captured or explicit old values),
keyframes_set_bezier_many (double precision), and
keyframe_set_bezier_point (single handle with NaN-capture fallback)
- dialog and curveview drag finalization go through the facade; the
tests exercise the same global undo stack via oakengine_project_undo
- keyframeview/keyframeviewundo.{h,cpp} removed with zero remaining
references
- new facade API: set_input_at_time (element addressing, track=-1 for
all components at once), set_input_string_at_time, frame_time_base,
array_insert_at/remove_at, disconnect_ex (element-aware), and
keyframes_set_type_many (first cross-track keyframe op, addressed by
(time,track) pairs)
- the widget bridge's commit funnel, color path, array ops, label
disconnect, and keyframe set-type actions in keyframeview/curvewidget
now go through the facade; keyframeviewundo.h loses two consumers
- deliberate leftovers with rationale: keyframecontrol's multi-track
composite ops (documented track-0-only limitation of the keyframe
family), keyframeproperties dialog (needs a set_time primitive),
curveview's drag UX, and NodeInputDragger (already engine-side)
- new primitives: ripple_delete_in_to_out (ripple or gap fill plus
work-area state, one undo command), trim_clips_to (batch edge trim
returning a count), delete_empty_tracks (type-filtered batch), and
marker_remove_many (sparse marker deletion by timestamp array)
- delete-in-to-out, edit-to, delete-all-empty-tracks, and sequence
viewer marker deletion now go through the facade; empty operations
no longer push empty undo entries
- footage viewer marker deletion keeps its app path deliberately
(facade marker handles are Sequences, not generic viewers); the
tentative subtitle track and pointer drag chain stay as documented
leftovers
- new batch primitives: split_clips (link-preserving, single undo
command), delete_clips (gap replace + optional ripple with explicit
region support), ripple_delete_range, marker_add_ex with color
- razor/split-at-playhead, clip delete, ripple-to-point, track delete,
and the non-dialog marker path now issue facade commands instead of
the app's own undo command classes
- batch operations deliberately produce one undo command per user
action (deleting twenty clips is one entry, not twenty); selection
and transition removal stay UI-side as documented leftovers
- new facade API: video params ex (dimensions/rate/par/interlacing/
preview format/divider), audio params, preview divider, and
undoable-flagged setters mirroring the dialog's dual undo/no-undo
modes; label setting gains node_set_label_ex
- the dialog's own SequenceParamCommand is gone; accept now issues
facade calls (per-field commands, unchanged fields skipped)
- preset system stays UI-side by design: its flat XML schema never
touches engine objects
- the auto-cache checkbox maps to the engine's existing stub (no undo
noise)
- new facade API: video stream overrides (colorspace/range/interlacing/
premultiply), pixel aspect, image-sequence params, stream enable,
source start time, and colorspace candidates - all undoable
- project explorer proxy actions now run through FacadeProxyTask and
the facade media-management functions (ProxyManager references in
projectexplorer.cpp drop from 5 call sites to a comment)
- footage properties dialog reads/writes through the facade; its two
app-side undo command classes are gone
- handle-model fix: the footage handle is a heap state object, not a
plain pointer cast - oakengine_footage_borrow() wraps app-held
Footage nodes correctly (nine UB reinterpret_casts caught by the
DialogFootageProperties tests)
- oakengine_export_render_ex covers the dialog's entire option surface
(formats, codecs, pix fmts, audio params, ranges incl. still frame,
subtitles, scaling, threads, custom OCIO color transform names,
per-codec key/value options) - zero feature reduction
- the dialog's Start now creates a FacadeExportTask that drives
oakengine_export_render_ex instead of constructing ExportTask in the
UI; progress flows through the facade callback and cancel through
oakengine_export_cancel (OAKENGINE_E_CANCELLED preserves the
keep-dialog-open semantics)
- two real fixes: audio sample format 0 no longer means an
AAC-unsupported u8_p (default is f32_p), and image-sequence exports
use the engine's real [#####] placeholder instead of a made-up -%04d
- clip-level loop mode (off/loop/clamp), undoable
- per-channel audio levels as one-frame RMS at a timestamp
- waveform min/max buckets over a footage range, rendered on demand
- two real fixes uncovered by this family: conform completion signals
were starved by msleep-only waits (audio renders always came back
empty), and incomplete tickets from conform-pending renders are now
retried until the conform is ready
- relink through Footage::set_filename with the engine's natural
clear+reprobe cascade; find_offline_footage recovers offline media
by exact filename match under a search directory
- synchronous proxy generate (event-loop driven like the export
family), state query, enable/disable, path, and delete, mirroring
the app's proxy dialog assembly
- documented as non-undoable, matching the app's existing semantics
- undoable track removal (content restored on undo) and true-move
track reordering assembled from the edge commands (the engine has
no move-track API)
- track height/mute/lock getters and setters, documented as
non-undoable to match the engine's current semantics
- undoable marker add/remove/rename; duplicate timestamps are
rejected with E_STATE instead of hitting the engine's debug assert
- keyframed state, count, read (time in sequence timebase + mapped
values), add/remove, easing read/write (linear/bezier/hold with
control points), and clear - all undoable with full undo/redo
assertions
- easing set commands are minimal capi-local UndoCommands matching
the app-layer semantics (engine has none of its own); same-time
duplicates are rejected with E_STATE instead of hitting the engine's
debug assert
- project node enumeration, type/name/label metadata, input
introspection (id, mapped value type, connection state)
- typed parameter read/write for the eight common NodeValue kinds
(int/float/bool/rational/color/vec2-4/combo/string), undoable via
NodeParamSetSplitStandardValueCommand - the split-track value path,
chosen after proving the standard-value command stores whole
variants into a single component track
- graph operations (add/remove/connect/disconnect) reusing the
engine's undo commands, all covered by undo/redo assertions
All four are undoable and reuse the timeline/ command classes verbatim
(BlockSplitCommand, TrackRippleRemoveAreaCommand, BlockTrimCommand,
ReplaceBlockWithGap+PlaceBlock), with undo/redo assertions covering
split halves, media-in alignment on trim, ripple shift amounts, and
full restoration on undo
- oakengine_export_render drives ExportTask synchronously (offline
render + encode) with a progress callback, codec probing, and a
thread-local error channel; exporter.h keeps clear of the visibility
macro header
- oak-cli transcode now defaults to mp4 (H.264/AAC) with --format ppm
keeping the raw output path
- two real concurrency bugs found by the facade's own test: ExportTask
deadlocks when start()ed synchronously (queued conform handshake
needs an event loop), and the progress callback must be captured by
value because it fires on the task thread
- oakengine_sequence_add_track and add_footage_clip are the facade's
first editing primitives: undoable track creation and clip placement
with full range validation, clip enumeration, and gap filtering
- oak-cli transcode closes the loop: media file -> import -> clip ->
render, producing scaled PPM frames and a WAV from just the C ABI
- engine fix uncovered by transcode: the render worker used an invalid
empty AudioParams for IPC render frames, crashing any sequence that
contains audio; it now derives them from the rendered node itself
- sequence_new hardens its defaults against missing audio config keys
- oakengine_footage_probe inspects media without a project (decoder
probe): stream counts, per-stream video info (dimensions, rate,
duration, color tags, interlacing), audio info, duration, decoder
name, source start time; thread_local last_error for the NULL-handle
failure paths
- oakengine_project_import_footage adds probed footage to a project
through the same undoable command path as the app
- oak-cli probe prints decoder/duration/per-stream details and runs as
a ctest everywhere (no GL)
- dual ownership documented: probe handles are owned, imported footage
is borrowed from its project
- oakengine_sequence_get_video_params (width/height/pixel aspect) with
assertions for the default 1920x1080 square-pixel sequence and the
invalid-handle path
- oak-cli now renders at the sequence's real dimensions instead of the
hardcoded 1920x1080
- oakengine_renderer_create/set_mode/last_error, render_frame (sync,
60s timeout, CPU frames via the existing worker pool),
render_audio (planar float), cancel; frames and audio buffers are
owned handles with borrowed data pointers
- output colorspace names map to OCIO display transforms, with
graceful fallback to reference-space output
- oakengine_renderer_test: parameter validation and error paths need
no GL and always run; render assertions gate on
DynamicRenderer backend availability and SKIP cleanly otherwise
- fix a facade bug found by its own test: wait_for_ticket leaked a
connected lambda capturing a stack reference, which a subsequent
cancelled ticket could fire into reused stack memory
- oakengine/export.h establishes the OAKENGINE_API visibility macros;
include/oakengine/ipc.h is the first pure-C surface (41 functions:
shm, frame slot pool, and the worker IPC messages as POD<->JSON
build/parse), implemented in engine/src/capi/
- the IPC implementations move to engine/src/oliveimpl (namespace
olive::engine::internal::ipc); engine/render/ipc/*.h are rebuilt as
same-name/same-API wrapper classes forwarding across the C boundary
- FrameSlotMeta is shared with the C header verbatim so the app/worker
wire format (v1) is bit-identical; static_asserts pin sizeof and
field offsets
- spscringbuffer.h moves to include/oakengine/ as an inline-only
header (no symbols, not ABI)
- new pure-C test oakengine_ipc_test (make_oakengine_test, no GL)
covers shm, frame pool, message round-trips and the layout asserts;
full gtest suite stays green (1986 tests)
Mike-Solar