An edit that moves lines into other frames composes on until a paragraph
starts in the same frame and place as before, or to the story's end, as
an_edit_that_moves_lines_composes_up_to_the_next_frame_the_text_starts_in
shows. The ROADMAP line no longer cites an issue about selection lag.
Every edit of a story composed it from its first paragraph, so a typed
character in a long threaded story cost as much as laying out the whole
story: 456 ms in a story threaded over 300 pages.
A composition that lays every paragraph once keeps its state before each
paragraph. After an edit, composing starts one paragraph above the first
changed one, in the kept state and with the lines above copied. It stops
at the first unchanged paragraph it reaches in the state kept for that
paragraph, and copies the lines from there on with their story offsets
moved by the length of the edit.
- The cache does this when only the story differs from its entry.
- Composed whole as before: stories with footnotes, endnotes, tables,
anchored objects, named lists, tracked deletions, balanced columns,
vertical justification other than Top, column rules, type on a path,
spanning or split paragraphs, paragraph borders or shading, and a
layout that goes back over a paragraph (keep options).
- An edit that moves lines into other frames is composed to the end of
the story.
In a 240-paragraph story a typed character composes 3 paragraphs
(240 before).
Docs: docs/architecture.md, docs/gaps.md (gap 20), ROADMAP.
Refs #2298
The 30 s timer ran file.recovery.save on the UI thread, and
designcraft_format::save() deflated the entire package — every linked placed
file included — into a Vec in RAM: 3.6 GB and 30-100 s of zlib inside
DesignApp::ui() per autosave on a 190-page catalogue with 4.2 GB of links.
- file.recovery.save {background: true} snapshots the dirty documents (Arc
clones; no bytes copied) and queues them on one "designcraft-recovery"
thread, skipping a document whose previous write is still in flight.
file.recovery.progress reports {written, busy, errors}; the app's timer uses
it and polls from ui_frame_end while a write is outstanding. The synchronous
command is unchanged for MCP, scripts and tests, and the web build still
writes inline (it has one thread).
- designcraft_format gains write(), streaming the package to any Write + Seek,
so a large save needs no second copy of the archive.
- A snapshot still queued or running when its document is saved, reverted or
closed is cancelled before its atomic rename; a write that lands after the
poll is discarded, so a crash cannot offer back a copy older than the saved
file. A panicking write is the job's error instead of ending the thread and
leaving the document stuck as running.
Tests: the_recovery_timer_writes_off_the_interface_thread;
recovery::a_background_save_writes_recovery_data_the_app_can_recover,
a_saved_document_is_not_written_in_the_background,
a_snapshot_whose_metadata_failed_is_still_the_documents_to_remove,
a_background_snapshot_whose_metadata_failed_is_still_the_documents_to_remove,
a_write_cancelled_midway_publishes_nothing, only_a_queued_write_is_cancelled,
a_panicking_recovery_write_is_an_error_and_the_writer_keeps_going;
format::writing_to_a_stream_makes_the_same_file_as_saving_in_memory.
Gates: cargo fmt --all --check; clippy -p designcraft-engine -p
designcraft-format -p designcraft-ui-egui --all-targets -D warnings;
cargo test -p designcraft-engine -p designcraft-format.
Part of #1923, split for review: leaving linked placed files out of snapshots
is a separate pull request on top of this one.
Conflicts:
- README.md: one WGPU_VALIDATION_INDIRECT_CALL row (this branch's wording).
- apps/designcraft/src/gpu.rs: main's only change is a test tweak in the backend-fallback tests, which now live in gpu/backend.rs; kept this branch's file and ported the tweak there (the check depends on the chosen backend, not the candidate count).
- apps/designcraft/src/main.rs: dropped main's inline VALIDATION_INDIRECT_CALL block (#664) in favour of gpu::instance_flags(); kept the adapter selection, restart and first-frame watch, and took main's macOS Dock Quit handling, preserve_documents_on_exit in on_exit, and profile::config_dir/recovery_dir.
- crates/engine/src/recovery.rs: took main's version; its retained recovered checkpoints do what this branch's recovery change did.
- ROADMAP.md: main's file plus this PR's one progress-log line.
Save and Save a Copy put a picture of the first page in the .designcraft
zip as preview.png (320 px on the longer side). The start screen reads
only that entry and shows the last eight recent files as cards with the
picture and the name, below the community card. Files saved before have
a blank page until they are saved again.
The package fallback-font test writes its fixture without Save's preview:
drawing it before the document's fonts load pulled in a system font that
covers the test's characters (Unifont CSUR) ahead of the document's own.
Conflict: Cargo.toml: next's winit patch (rust-windowing/winit#4694, Windows mixed-DPI drags) kept; main's skylartaylor/winit patch held back (Cargo takes one patch per crate; both are single commits on v0.30.13 for different platforms, no published commit has both)
Conflict: Cargo.lock: next's winit source; main's ui-egui tempfile dev-dependency taken
Conflict: apps/designcraft/src/main.rs: next's startup; main's macOS with_close_requested_on_terminate block left out (needs main's winit patch); apps/designcraft/Cargo.toml keeps next's (no macOS winit dependency)
Conflict: ROADMAP.md: headers combined, logs both
Conflict: docs/target-app-parity.md: headers combined
Route AppKit termination through the existing unsaved-document close transaction using the immutable winit patch. Restore pending prompts from hidden or minimized windows and preserve answers across repeated requests.
Validation: all six local gates, native cancellation/retry/save smoke, and hosted platform checks.
Fixes#544.
PR #623.
upstream/main c3859c82 (#614e198eb5d, #6159509b2bc, #6168e151a2f, #617c3859c82).
Conflict: ROADMAP.md: header keeps next's change list plus orderly native exit recovery; progress log both
Conflict: apps/designcraft-cli/src/main.rs: next's case-insensitive .jpg/.jpeg check with main's write_atomic
Conflict: apps/designcraft/src/main.rs: on_exit keeps next's startup.presented() (not an Option) and adds preserve_documents_on_exit; next's discard_marked_text and main's preserve_documents_on_exit both; Services write uses write_atomic and keeps next's clipboard_text; tests both
Conflict: crates/engine/src/cmd/book.rs: main's ChapterBatch paginate with next's chapter-number resolution and {pages, chapters} result and doc
Conflict: crates/engine/src/cmd/file.rs: revert takes main's discard_document ordering without the revision + 1 (next's DocState::new hands out a fresh revision)
Conflict: crates/engine/src/recovery.rs: main's save_tracked, write_entry and open (st.bump_revision() for next's global revisions); next's reopen-twice test assertions kept
Conflict: docs/gaps.md, docs/target-app-parity.md: headers combined; next's Language, Books and Numbering lines plus main's book update safety line
Checkpoint the latest document state during App::on_exit after honoring current quit approvals. Keep incomplete or stale plans, clear recovery for clean documents and report unavailable or failed storage.
Follow-up to #545; cancelable macOS Dock Quit remains open in #544.
PR #617.
Ported from VectorCraft's start-up fixes (#502, #545, #651, the display-GPU
choice, #806, #964) and reconciled with the backend fallback of #297:
- The app picks the window's adapter itself (eframe's native_adapter_selector,
apps/designcraft/src/gpu.rs): adapters that can't present never, then
WGPU_ADAPTER_NAME, hardware before software, native backends before OpenGL,
the GPU that drives the primary display (Windows EnumDisplayDevices through
winsafe, Linux /sys/class/drm), the power preference (WGPU_POWER_PREF;
otherwise power saving on Windows and macOS, the system's order elsewhere),
then DX12 before Vulkan. Every adapter and the choice are logged.
- wgpu's indirect-call validation is off (WGPU_VALIDATION_INDIRECT_CALL=1 turns
it back on): its compute shader failed to compile on an Intel Mac and lost
the Metal device (#334, first error).
- run_native runs under catch_unwind. A wgpu error or a panic in wgpu,
egui-wgpu or naga during the first frames, or a first frame that doesn't
reach the screen in 20 s, starts the app again without that adapter
(DESIGNCRAFT_GPU_SKIP; exec on Unix, a new process on Windows). When no
adapter of the start's backends is left, it starts again with the next
backend. Each restart leaves out one more adapter or backend, so they end.
Adapters without PCI ids (Metal) are keyed by name.
- The backend fallback (gpu/backend.rs, gpu.json) stays for driver faults that
take the process down. Windows tries DirectX 12, then OpenGL, then Vulkan,
each in its own instance: Vulkan last because creating its instance loads
every installed Vulkan driver, and OpenGL kept out of the DX12 instance
because creating it crashed in AMD's atio6axx.dll (#167).
- main returns an ExitCode and logs why the app stopped to standard error and
the log file, which is all a Windows start without a console leaves (#273).
README, architecture, gaps, hardware and target-app parity docs and ROADMAP
describe the choice, the restarts and the variables.
Progress docs per craftrules standards/progress-docs.md: target-app parity, gaps, roadmap with alpha gate, readiness by audience (full / mainstream / essentials) with Opus 5.5 hours, and feature, format, hardware, UI and localization parity docs.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
dotCooCoo
francesco fantoni
aiongg
skylar taylor-barrick
PiMinusMax
Brandon ThomasandClaude Opus 5.5