Files
likang 7e01147566
CI / typecheck + test + build (push) Has been cancelled
Initial commit: import VibeMake PCB Studio
2026-11-03 11:20:08 +08:00

75 KiB
Raw Permalink Blame History

eeschema — divergences from KiCad

Reference: /home/akshay/kicad-reference/eeschema (10.0.6 as of 2026-09-29; was 10.0.5 through most of this file's history — a separate pass ports the 10.0.5→10.0.6 delta for code already ported, this file's own root-path findings below were re-checked against 10.0.6 and did not move). Layout rule: same as pcbnew/STRUCTURE.md — every module sits at the path KiCad keeps its counterpart at, and there is no src/.

Only divergences are listed. Anything not mentioned matches. Add a row when you find one; keep the reason to a line. Regenerate the raw classification with qa/probes/struct_diff_eeschema.sh (writes docs/eeschema-structure-diff.md) before trusting this file's counts — this file is the curated, content-checked read of that probe's output, not a transcription of it: the probe matches by filename only, so several of its MOVED/ELSEWHERE hits below are noted as false positives rather than acted on.

Parity, 2026-10-10 (file level)

Regenerate with python3 qa/probes/eeschema_parity.py. Counting (decided 2026-10-10): a KiCad .cpp and its header are one file; a header counts alone only with no .cpp beside it and code in it (declaration-only invoke_sch_dialog.h / save_project_utils.h do not); a wxFormBuilder _base folds into its dialog, and counts as the dialog when none derives from it; our _ui.tsx is the same file as its .ts. Folders with nothing to match (n/a, deferred, sim) are not shown. A file present is not a file complete - GetMsgPanelInfo was missing from 17 classes whose files count as done - so read the "left" column as a floor. Decisions behind the columns:

  • n/a (cannot exist in a browser): api/ (IPC/protobuf, no process boundary), navlib/ (3Dconnexion SpaceMouse driver, as pcbnew/STRUCTURE.md), the two native file-dialog hooks (widgets/filedlg_hook_save_project, widgets/symbol_library_save_as_filedlg_hook) and symbol_chooser_timing (a debug timing header).
  • deferred: the third-party sch_io importers (formats we did not read; the pcbnew session ports them on eeschema-dev now) and KiCad jobsets (eeschema_jobs_handler, dialog_erc_job_config - pcbnew skips its job dialogs the same way).
  • sim: deferred (decided 2026-10-10): sim/, sim/kibis/ and the simulator's dialogs, preference panel, simulator_control tool and tuner widget. It needs ngspice built for the browser first.

Table format (keep it): one row per folder - folder | KiCad | n/a | deferred | sim | to match | done | left | done % | less in ours | extra in ours | extra kept | extra record model | extra unexplained - then a total row. less in ours is what KiCad has and we lack (the files still to port); extra in ours is every file we have that KiCad does not, split in the last three columns by why it exists. The probe prints exactly this.

folder KiCad n/a deferred sim to match done left done % less in ours extra in ours extra kept extra record model extra unexplained
(root) 94 1 1 0 92 91 1 98% 1 12 0 12 0
browser 0 0 0 0 0 0 0 - 0 6 5 1 0
connectivity 0 0 0 0 0 0 0 - 0 6 6 0 0
dialogs 63 0 1 6 56 33 23 58% 23 2 0 0 2
erc 5 0 0 0 5 5 0 100% 0 1 0 0 1
exporters 0 0 0 0 0 0 0 - 0 1 1 0 0
import_gfx 3 0 0 0 3 3 0 100% 0 2 0 1 1
libraries 2 0 0 0 2 1 1 50% 1 0 0 0 0
netlist_exporters 11 0 0 0 11 10 1 90% 1 0 0 0 0
printing 2 0 0 0 2 2 0 100% 0 0 0 0 0
sch_io 3 0 0 0 3 3 0 100% 0 0 0 0 0
sch_io/kicad_legacy 3 0 0 0 3 3 0 100% 0 0 0 0 0
sch_io/kicad_sexpr 4 0 0 0 4 4 0 100% 0 0 0 0 0
sch_io/sexpr 0 0 0 0 0 0 0 - 0 3 3 0 0
sim 60 0 0 60 0 0 0 - 0 1 0 0 1
symbol_editor 11 0 0 0 11 5 6 45% 6 9 0 5 4
sync_sheet_pin 7 0 0 0 7 7 0 100% 0 0 0 0 0
tools 30 0 0 1 29 24 5 82% 5 36 0 33 3
widgets 18 2 0 1 15 8 7 53% 7 1 0 0 1
total 343 7 23 70 243 199 44 81% 44 80 15 52 13

Left to port

  • (root) (1): bom_plugins
  • dialogs (23): dialog_bom, dialog_bom_help_md, dialog_database_lib_settings, dialog_import_symbol_select, dialog_lib_edit_pin_table, dialog_lib_fields_table, dialog_lib_new_symbol, dialog_lib_symbol_properties, dialog_migrate_buses, dialog_pin_properties, dialog_remote_symbol_config, dialog_rescue_each, dialog_symbol_chooser, dialog_symbol_remap, dialog_update_symbol_fields, panel_eeschema_color_settings, panel_eeschema_display_options, panel_eeschema_editing_options, panel_sch_data_sources, panel_sym_color_settings, panel_sym_display_options, panel_sym_editing_options, panel_sym_lib_table
  • libraries (1): legacy_symbol_library
  • netlist_exporters (1): netlist
  • symbol_editor (6): lib_logger, lib_symbol_library_manager, symbol_editor_import_export, symbol_editor_plotter, symbol_editor_undo_redo, symbol_saveas_type
  • tools (5): symbol_editor_control, symbol_editor_drawing_tools, symbol_editor_edit_tool, symbol_editor_move_tool, symbol_editor_pin_tool
  • widgets (7): hierarchy_pane, panel_remote_symbol, pinshape_combobox, pintype_combobox, symbol_filter_combobox, symbol_preview_widget, symbol_tree_pane

Extra: old record model (deleted as each consumer moves to the live model)

browser/eeschema_app, fieldbox, hover_selection, import_gfx/graphics_importer_sch_mapping, index, lib_tree_item, net_overrides, project, project_settings, project_sym_lib_table, sch_record_bridge, sch_script_api, symbol_editor/conditions, symbol_editor/edits, symbol_editor/symbol_editor_dialogs, symbol_editor/symbol_renderer, symbol_editor/toggles, toggles, tools/bbox, tools/build, tools/build-graphics, tools/cleanup, tools/clipboard, tools/command, tools/connect, tools/directive_label, tools/edit_symbol_libid, tools/hittest, tools/hop_over, tools/image_size, tools/intersheet_refs, tools/label_properties, tools/move, tools/mutate, tools/page_settings, tools/pin_alternates, tools/point_editor, tools/properties, tools/rule_area, tools/save_symbol_to_schematic, tools/scene_bbox, tools/sch_align_record, tools/sch_request_selection, tools/sch_selection_filter, tools/sch_sheet_pin_tool, tools/symbol_from_schematic, tools/symbol_unit, tools/table_cells, tools/table_layout, tools/transform, types

Extra: no recorded reason yet (fold, rename or justify)

dialogs/item_color, dialogs/panel_setup_severities, erc/marker_nav, import_gfx/image_format, sim/sim_model_types, symbol_editor/cursors, symbol_editor/defaults, symbol_editor/grid, symbol_editor/symbol_edit_frame_app, tools/arc_edit, tools/assign_netclass, tools/index, widgets/net_navigator_panel

The 16 unexplained extras, 2026-10-10: where each belongs, for a decision

file (ours) KiCad home proposal
dialogs/dialog_increment_annotations dialogs/dialog_increment_annotations_base.cpp - upstream has no derived class; SCH_EDITOR_CONTROL::IncrementAnnotations builds the _BASE dialog itself (sch_editor_control.cpp:2917) rename to dialog_increment_annotations_base, as upstream
dialogs/item_color none: our record colour <-> COLOR4D conversion delete with the record model
dialogs/panel_setup_severities common/dialogs/panel_setup_severities.cpp (shared) fold into the common panel (pcbnew uses it already)
erc/marker_nav RC_TREE_MODEL::PrevMarker / NextMarker (common/rc_item.cpp) move to common/rc_item.ts
import_gfx/image_format none: guesses the format of an embedded bitmap's bytes, which wxImage does upstream keep as a browser helper (browser/)
sim/sim_model_types sim/sim_model.cpp (SIM_MODEL::ReadTypeFromFields / InferSimModel) goes with the sim port (deferred); ERC's TestSimModelIssues uses it until then
symbol_editor/cursors none: the cursor CSS a tool shows fold into the tools' SetCurrentCursor, as eeschema's canvas does
symbol_editor/defaults SYMBOL_EDITOR_SETTINGS::m_Defaults read where an item is made delete with the symbol editor's live switch
symbol_editor/grid EDA_DRAW_FRAME::GetNearestGridPosition on the GAL grid delete with the symbol editor's live switch
symbol_editor/symbol_edit_frame_app none: what the program gives the window, as browser/eeschema_app move to browser/, as eeschema_app
tools/arc_edit enum class ARC_EDIT_MODE in include/settings/app_settings.h move the enum to common/settings/app_settings.ts
tools/assign_netclass the getNetNamePattern lambda of SCH_EDITOR_CONTROL::AssignNetclass (sch_editor_control.cpp:1183) fold into sch_editor_control.ts with the live AssignNetclass
tools/index none: a barrel delete when the record tools it re-exports are gone
widgets/net_navigator_panel SCH_EDIT_FRAME::RefreshNetNavigator (net_navigator.cpp) and the frame's wxTreeCtrl the view half of net_navigator.ts; rename net_navigator_ui.tsx
sch_canvas the canvas half of SCH_EDIT_FRAME's constructor see the root table
toggles frame settings read by EDITOR_CONDITIONS see the root table

Root, 2026-10-10: what is left, and what is waiting on a decision

Ported this pass: general.ts, sch_text_help_md.ts, sch_preview_panel.ts. Moved: eeschema_app.ts -> browser/ (pcbnew's pcbnew_app.ts precedent). The root's remaining differences, each with the reason it is not done, for a decision:

item KiCad why not matched proposal
invoke_sch_dialog.h header only: declares the Invoke* dialog launchers, each defined in its dialog's .cpp a TS module cannot declare what another module defines; the launchers live with their dialogs as upstream's definitions do n/a (header-only declarations)
save_project_utils.h header only: declares PrepareSaveAsFiles, defined in files-io.cpp ours is defined in files-io.ts, as upstream; the header adds nothing to port n/a (header-only declaration)
bom_plugins.cpp BOM_GENERATOR_HANDLER: runs an external BOM generator (xsltproc / python) on the netlist a browser runs no external process n/a, or deferred with dialog_bom
index.ts (ours) none: pcbnew has no package barrel 154 importers use it as the package entry keep as the package entry, or replace by direct imports
sch_canvas.ts (ours) the canvas half of SCH_EDIT_FRAME::SCH_EDIT_FRAME the DOM element comes from the window; this is the frame's side, as pcbnew's pcb_canvas.ts fold into sch_edit_frame.ts, or browser/
toggles.ts (ours) the left toolbar's toggle state, upstream frame settings read by EDITOR_CONDITIONS window state the toolbar renders from port onto the frame's settings and conditions, then delete
types.ts, sch_record_bridge.ts, hover_selection.ts, net_overrides.ts, sch_script_api.ts (ours) none: the old record model consumers not all moved to the live model yet delete with S7, each with its last consumer

Root match, before and after (stage E1, 2026-09-28)

KiCad's eeschema/ root has 83 .cpp files. Ours had 12 files total, 6 of which matched a KiCad root name (lib_symbol.ts, pin_type.ts, schematic.ts, sch_field.ts, sch_pin.ts, sch_symbol.ts). Most of our code lived instead in eeschema/tools/ (86 files, our own names) and designer/src/editors/schematic/ (UI + logic mixed together).

After E1: eeschema root has 30 files, 23 of which are 1:1 KiCad root-name matches (up from 6/12). eeschema/tools/ is down from 86 to 76 files (11 moved out to the eeschema root/erc//widgets/, 1 moved in from designer/) — all either an exact tools/<name>.cpp match or a file with no KiCad counterpart (see below). 42 files total now sit at KiCad's own path, up from 18 before this stage (qa/probes/struct_diff_eeschema.sh's SAME count).

After E2 pt 1 (dialogs/widgets/render/symbols, same day): eeschema root has 45 files, 28 of which are 1:1 KiCad root-name matches. The probe's SAME count is 72 (up from 42), DIALOG is down to 5 and MOVED to 2 — most of what it used to flag as a designer/-only dialog now lives at KiCad's own dialogs/ path.

After E1 pt 2 (fused-file splits, same day, six moves — see below): eeschema root is 67 files, 49 of which are 1:1 KiCad root-name matches — this stage's own contribution is 6 (bus-wire-junction.ts, sch_reference_list.ts, sch_item_alignment.ts, pin_layout_cache.ts, symb_transforms_utils.ts, project_sch.ts); the rest of the root's growth since E2 pt 1 is concurrent work in this same checkout (other agents' SCH_* item-class ports), not this stage's. The probe's SAME count is 95 (up from 72).

What moved

Twenty-five renames/moves, each verified against the C++ doc comment already in the file (every ported module here carries a "Counterpart:" line) before moving, then re-typechecked (eeschema, designer, qa) and the affected tests re-run.

eeschema/tools/ → eeschema root (KiCad has these as root .cpp, not under tools/): fields_data_model.ts, annotate.ts, net_navigator.ts, junction_helpers.ts, autoplace_fields.ts, refdes_tracker.ts, sch_collectors.ts, sch_sheet_path.ts, project_rescue.ts. Plus cross_probe.ts → cross-probing.ts (KiCad's file is hyphenated).

Renamed in place (eeschema/tools/, KiCad's own tools/ name differs from ours): back_annotate.ts → backannotate.ts.

eeschema/tools/ → eeschema/widgets/: sch_properties_panel.ts (KiCad: widgets/sch_properties_panel.cpp).

eeschema/connectivity/ → eeschema/erc/: erc.ts (KiCad: erc/erc.cpp; connectivity/ was never a KiCad directory name here — see "Directories we have that KiCad doesn't").

eeschema/exporters/ → eeschema/netlist_exporters/ (KiCad has no exporters/, only netlist_exporters/): netlist_exporter_allegro.ts, netlist_exporter_kicad.ts, and spice.ts → netlist_exporter_spice.ts (KiCad: netlist_exporter_spice.cpp). netlist.ts (which fused netlist_exporter_xml.cpp + netlist_exporter_orcadpcb2.cpp under our own name) was itself split in stage F — see below; it no longer exists.

designer/src/editors/schematic/ → eeschema/ (plain .ts, zero designer/ imports, one clean KiCad counterpart apiece): schematic_settings.ts, generate_alias_info.ts, menubar.ts, sch_edit_frame.ts (the KIWAY_PLAYER shell half of SCH_EDIT_FRAME, same pattern as pcbnew/footprint_edit_frame.ts), toolbars_sch_editor.ts, files_io.ts → files-io.ts (KiCad's file is hyphenated), symbol_props_rows.ts → fields_grid_table.ts (KiCad: fields_grid_table.cpp / FIELDS_GRID_TABLE), symbol_search_terms.ts (no KiCad file of its own — it splits LIB_SYMBOL::cacheSearchTerms/cacheChooserFields out of lib_symbol.cpp for testability; moved beside lib_symbol.ts, its counterpart's actual home). sch_navigate_tool.ts → eeschema/tools/ (KiCad: tools/sch_navigate_tool.cpp).

Each move needed its own imports fixed by hand once: a file that crossed into the eeschema package still had @ziroeda/eeschema-prefixed imports for symbols now in its own package (self-imports don't resolve here), so fields_grid_table.ts, generate_alias_info.ts, schematic_settings.ts, symbol_search_terms.ts and tools/sch_navigate_tool.ts each got that one import turned relative. move_ts.py only rewrites references to the file it is moving, not a moved file's own pre-existing imports of its new package's other symbols.

String-literal path references (not import specifiers, so the mover's rewrite pass never sees them) needed a manual sweep and fix in nine qa/ test files and three source doc comments — see the commits for the list; the menu_ellipsis.test.ts allowlist entry for sch_collectors.ts was the one case where missing this would have silently dropped a pinned exception rather than fail loudly.

Stage E1 pt 2 (2026-09-28): the fused files split out

Six of the E1 task list's "kept fused" targets got their own file this stage, each a plain move (export the shared private helper the other side still needs, import it back) with the same tests re-run green:

  • bus-wire-junction.ts: SCH_EDIT_FRAME::TrimWire (out of tools/post_move_cleanup.ts) and DeleteJunction, as dissolveJunctionsAt (out of tools/cleanup.ts, which now exports mergeOverlap/mergedLine/ sameLayer for it to share). TestDanglingEnds/UpdateHopOveredWires stay view-side in designer/.../render/renderer.ts, unmoved.
  • sch_reference_list.ts: the numbering core — splitReference, annotateSymbols/annotateHierarchy, checkAnnotation, and the ANNOTATE_ORDER_T/ALGO_T/SCOPE_T enums the header declares — out of annotate.ts, which keeps the SCH_EDIT_FRAME side (the undoable commands, Increment Annotations, the two report loops) and imports the core back.
  • sch_item_alignment.ts: the pure grid-snap geometry (EE_GRID_HELPER::AlignGrid, the shifts histogram) out of tools/align_to_grid.ts, which keeps the per-item-type dispatch and the SCH_MOVE_TOOL::AlignToGrid commit/drag loop — still one call spanning both upstream files, so that half stays fused by design. MoveSchematicItem has no separate port (tools/move.ts/tools/connect.ts already do that dispatch).
  • pin_layout_cache.ts (move_ts.py rename from pin_box.ts): libPinBoundingBox/altIconBox are GetPinBoundingBox/GetAltIconBBox. The render-time text-layout half (GetPinNameInfo/GetPinNumberInfo/ SetRenderParameters/the caches) still has no port and stays out.
  • symb_transforms_utils.ts: pinSpinStyle (GetPinSpinStyle) out of tools/label_properties.ts. OrientAndMirrorSymbolItems/ RotateAndMirrorPin are still not ported — nothing calls them.
  • project_sch.ts: legacySchLibs (PROJECT_SCH::LegacySchLibs) out of SchematicEditor.tsx's legacyCache useCallback, as a pure function taking the file list and project root name as plain arguments (a structural SchLibFile shape, not designer's PickedFile, so the function has no UI-layer dependency). SchSearchS (no filesystem search path in a browser) and SymbolLibAdapter (library ids resolve through the ordinary loadSymbol/repairSourceLibs path, no separate cached adapter) are not ported.

Root files ported since the table above (2026-09-28/29)

sch_validators.ts, multiline_pin_text.ts, gfx_import_utils.ts, symbol_checker.ts, symbol_import_manager.ts, eeschema_helpers.ts and lib_fields_data_model.ts now exist at KiCad's own root paths. Each is a real, tested port, but most have no live caller yet (documented in the file's own doc comment) — that's a UI-wiring gap, not a missing port:

  • sch_validators.ts (SCH_NETNAME_VALIDATOR): ported onto common/validators.ts's new NETNAME_VALIDATOR. Still no text field calls it for live bus-name feedback.
  • multiline_pin_text.ts (ComputeMultiLinePinNumberLayout): ported whole. Still nothing produces the brace-wrapped multi-line pin-number string it lays out.
  • gfx_import_utils.ts (ConvertImageToPolygons/ConvertImageToLibShapes): ported on the shared WX_IMAGE. ConvertSVGToLibShapes is not ported — undeclared in the upstream header and callerless there too.
  • symbol_checker.ts (CheckLibSymbol/CheckDuplicatePins/ CheckLibSymbolGraphics): ported whole, reusing LIB_SYMBOL.GetLogicalPins instead of re-walking GetStackedPinNumbers. The Symbol Editor's live checker dialog (designer/src/editors/symbol/components/dialogs.tsx's checkLibSymbol) is a separate, already-working port of the same C++ against that editor's own UI-side LibSymbol shape — not merged in, since rewriting the live dialog onto LIB_SYMBOL is a larger, separate change.
  • symbol_import_manager.ts (SYMBOL_IMPORT_MANAGER): ported whole — upstream's own header calls it "designed to be UI-independent for testability", and it is here too (no LIB_SYMBOL method is called, only stored).
  • eeschema_helpers.ts (EESCHEMA_HELPERS): only the two pure pieces of LoadSchematic — chooseSchFileFormat (format-by-extension dispatch) and resolveRootSheetName (root-sheet display name from top_level_sheets). The headless load pipeline itself (SETTINGS_MANAGER/ PROJECT resolution, the SCH_IO parse, SCH_SHEET_LIST/SCH_SCREENS migration, TOOL_MANAGER/SCH_COMMIT wiring for RecalculateConnections) is still not built — see the row below.
  • lib_fields_data_model.ts (LIB_FIELDS_EDITOR_GRID_DATA_MODEL): the column model, data store, cell read/write, grouping, sorting and the checkbox attribute getters/setters — the pure data-model half fields_data_model.ts/FieldsDataModel already draws the same line at. Not ported: the WX_GRID notifications (view-only), ExpandRow/ CollapseRow/CollapseForSort/ExpandAfterSort (UI row-visibility bookkeeping), ApplyData (writes the store back onto the library's LIB_SYMBOLs) and CreateDerivedSymbol/createActualDerivedSymbol (creates a new derived LIB_SYMBOL in the library) — all four need a live library-mode Symbol Fields Table window, which does not exist here (only the schematic-mode dialog does).

Root/SAME counts above predate this batch and were not regenerated (qa/probes/struct_diff_eeschema.sh) — read them as of 2026-09-28, not current.

Not ported — no file to move

KiCad file status
eeschema_settings.cpp (EESCHEMA_SETTINGS, eeschema.json) The type+defaults half moved to eeschema/eeschema_settings.ts (EeschemaSettings/EESCHEMA_DEFAULTS, out of designer/src/prefs/settings.ts) 2026-09-28. No EESCHEMA_SETTINGS extends JSON_SETTINGS class exists, and none is planned: the precedent this tree already set for a per-app settings file (pagelayout_editor/pl_editor_settings.ts's PL_EDITOR_SETTINGS) only wraps the handful of fields genuinely unique to that C++ class in a small FromJson/ToJson pair; upstream EESCHEMA_SETTINGS has no such small "own fields" set (it's struct-of-structs, APPEARANCE/AUI_PANELS/AUTOPLACE_FIELDS/…, all inherited-looking groups), and the app's actual runtime store is designer/src/prefs/settings.ts's one combined JSON, not a per-frame file a JSON_SETTINGS subclass would load/save on its own — building one nothing calls would duplicate EeschemaSettings, not replace it.
eeschema_helpers.cpp (EESCHEMA_HELPERS) The headless load pipeline itself — SETTINGS_MANAGER/PROJECT resolution, the SCH_IO plugin parse, SCH_SHEET_LIST/SCH_SCREENS migration, TOOL_MANAGER/SCH_COMMIT wiring for RecalculateConnections — is not built: no CLI or process boundary for it to serve, and it would reach into SCHEMATIC/connection_graph/sch_io/kicad_sexpr, mid-port elsewhere in this tree as of 2026-09-29. Its two pure pieces are ported; see the table above.
lib_fields_data_model.cpp/.h (LIB_FIELDS_EDITOR_GRID_DATA_MODEL) The pure data-model half is ported (see the table above); the wxGrid/library-write-back/derived-symbol-creation half needs a live library-mode Symbol Fields Table window, which does not exist here.

Own-named root files: fold audit (2026-09-28)

Checked each against its cited C++ counterpart for a clean rename/merge target. Only pin_box.ts had one (above, → pin_layout_cache.ts); the rest have no safe fold:

ours cited counterpart why it stays
project_settings.ts 4 different files (schematic_settings.cpp, erc/erc_settings.cpp, common/project/net_settings.cpp, common/project/project_file.cpp) No single KiCad file to fold into — it is our own project-file-persistence aggregate over four namespaces.
render_color.ts, symbol_markers.ts, pin_alt_icon.ts SCH_PAINTER methods (eeschema/sch_painter.cpp) No eeschema/sch_painter.ts exists to fold into — rendering lives split across designer/.../render/*.ts per this codebase's own architecture, not as one ported painter class.
pdf_annotations.ts SCH_SHEET::Plot/SCH_SYMBOL::Plot (the item classes E3 owns) via sch_plotter.cpp No sch_plotter.ts exists either; the two Plot() methods it mirrors live inside the busy SCH_* item files this stage does not touch.
global_sym_lib_table.ts template/sym-lib-table (an installed data file, no .cpp) Data, not code — nothing to fold into.
project_sym_lib_table.ts cites eeschema/symbol_lib_table.cpp That file does not exist in this 10.0.5 tree at all — SYMBOL_LIB_TABLE appears to have been superseded by SYMBOL_LIBRARY_ADAPTER (eeschema/libraries/symbol_library_adapter.cpp, see project_sch.h's PROJECT_SCH::SymbolLibAdapter). Renaming without verifying behavioural equivalence against the new class is out of scope for a moves-only stage; flagged here instead.
repair_source.ts DIALOG_CHANGE_SYMBOLS::processSymbols Supports a dialog already ported as tools/change_symbols.ts (see the dialog-adjacent-logic table below); it is not itself a .cpp file to fold into.
lib_tree_item.ts include/lib_tree_item.h Already matches (the header itself is not under eeschema/ upstream either). No move.
default_values.ts eeschema/default_values.h Already matches. No move.
fieldbox.ts inline in SCH_PAINTER::draw(SCH_FIELD) (sch_painter.cpp) No dedicated KiCad file — there never was one to fold into.
symbol_search_terms.ts LIB_SYMBOL::cacheSearchTerms/cacheChooserFields (lib_symbol.cpp) lib_symbol.ts exists and is the right home, but this split was deliberate (own doc comment: "for testability") and has only 3 importers either way — folding it in was considered and declined rather than undoing a stated design choice mid-stage.
lib_symbol_compare.ts LIB_SYMBOL::Compare (lib_symbol.cpp) Same target file, but 30+ importers across tools/, connectivity/, erc/, netlist_exporters/ — physically merging it into lib_symbol.ts is a large-blast-radius rename across files several of which are mid-edit by other agents in this checkout. Declined for this stage on risk, not on the merits.

Directories KiCad has that we don't

Same reasoning as pcbnew/STRUCTURE.md's table: api/ (IPC/protobuf, no process boundary here), python/ (SWIG), no interpreter. Import plugins for formats we don't read (sch_io/altium, geda, ltspice, eagle, easyedapro, pads, http_lib, easyeda, cadstar, database) — we only port kicad_legacy (as sch_io/legacy/) and kicad_sexpr (as sch_io/sexpr/). microwave/, specctra_import_export/ equivalents don't apply to eeschema. dialogs/, printing/ dialog bodies, and most of widgets/ stay in designer/src/editors/schematic/{dialogs,widgets}/ for stage E2 (UI windows) — see that stage's own note when it lands.

Directories we have that KiCad doesn't

connectivity/ (bus.ts, dangling.ts, hierarchy.ts, nets.ts, segment_index.ts) — no matching KiCad directory; our netlist-building graph walk where KiCad's is one file, connection_graph.cpp (2833 lines, unported as a single module — see eeschema-netlist-parity-backlog.md). exporters/ — down to just bom.ts after this stage (the netlist exporters moved out to netlist_exporters/); bom.ts has no KiCad file of its own (the Symbol Fields Table's BOM export is inside dialogs/dialog_symbol_fields_table.cpp, no separate translation unit), so it stays under our invented name.

Content divergences / false-positive probe matches

The blunt filename-only probe (struct_diff_eeschema.sh) flags these as MOVED/ELSEWHERE, but reading the code shows a different KiCad function under the same basename, so no move was made:

ours probe's guess actual counterpart
project.ts common/project.cpp (PROJECT, the app-wide project singleton) Our own SCH_SHEET_LIST-style multi-sheet hierarchy walk — unrelated code, same word.
tools/clipboard.ts common/clipboard.cpp SCH_EDITOR_CONTROL's copy/paste (eeschema/tools/sch_editor_control.cpp) — one slice of a large god-tool-class we split by action, not a generic clipboard helper.
tools/transform.ts libs/kimath/src/transform.cpp (a 2D matrix type) SCH_EDIT_TOOL::Rotate/::Mirror (eeschema/tools/sch_edit_tool.cpp) — unrelated code, same word.

eeschema/tools/ — kept under our own name (76 files, no KiCad tools/<name>.cpp)

KiCad decomposes schematic editing into a handful of large tool classes (SCH_EDIT_TOOL, SCH_EDITOR_CONTROL, SCH_DRAWING_TOOLS, SCH_MOVE_TOOL, SYMBOL_EDITOR_*) each thousands of lines; we split by action/concern instead, one small file per behaviour (move.ts, connect.ts, ortho.ts, hop_over.ts, swap_pins.ts, unfold_bus.ts, table_*.ts, and so on — the full list is the OURS section of docs/eeschema-structure-diff.md). None of these has a 1:1 KiCad filename, by construction, so none moves. This mirrors pcbnew's own tools/ split and is not itself a divergence to fix.

Dialog-adjacent logic, kept in tools/ (not moved to designer/)

exporters/bom.ts, tools/change_symbols.ts, tools/field_properties.ts, tools/global_edit_text_and_graphics.ts, tools/label_properties.ts each match a KiCad dialogs/dialog_*_base.cpp by name (the probe's DIALOG bucket), because upstream fuses the dialog window and its model logic into one class. We split them, so there is no non-dialog KiCad file to rename these to — dialogs/, printing/dialog_print.* etc. are UI-window targets for stage E2, not this stage's erc//sch_io//netlist_exporters// printing//widgets/ bucket. They stay in tools/ as the engine half of each dialog, same architecture as the rest of tools/.

Non-UI designer/src/editors/schematic/ modules left in place

These have zero designer/ imports (checked directly), same test as the files that did move, but no single KiCad file to move them under — either the KiCad code is a method inside a much larger class file whose "real" name is already claimed by something else we moved (sch_edit_frame.cpp → sch_edit_frame.ts, menubar.cpp → menubar.ts), or it has no upstream file of its own at all:

cursors.ts, hotkeys.ts, hotkey_list.ts, hotkey_bindings.ts, hover_selection.ts, moving_ids.ts, panes.ts, pin_icons.ts, toggles.ts, theme.ts — all say so in their own doc comments (fused into sch_actions.cpp/common/hotkeys_basic.cpp/sch_selection_tool.cpp/ dialog_symbol_properties.cpp/eeschema_settings.cpp, or no upstream file: moving_ids.ts, toggles.ts). back_annotate_source.ts (the board-read half of BACK_ANNOTATE, complementing tools/backannotate.ts) and frame_title.ts (SCH_EDIT_FRAME::updateTitle, one method of the class sch_edit_frame.ts already claims) are the two closest to moving, but splitting either out from its class further is the same refactor risk as the "kept under our own name" table above. net_overrides.ts (SCH_LINE/SCH_JUNCTION methods, fused into those item classes), and template_fieldnames.ts (common/template_fieldnames.cpp — a common/ candidate, out of scope for an eeschema-only stage) round out the list. (project_settings.ts moved in stage E2 pt 1 below — it turned out to have no designer/ imports of its own, just no single KiCad file.)

UI windows (stage E2)

Stage A-style siblings, pt 1: dialogs, widgets, render/symbol fragments (2026-09-28)

Same method as pcbnew/STRUCTURE.md's "Frame chrome" stage A: every designer/src/editors/schematic/{dialogs,components,render,symbols} module SchematicEditor.tsx names by relative import, that itself has zero designer/ imports (checked transitively, not just one hop — two modules that looked clean on a same-tree-prefix check turned out to reach a genuinely designer/-coupled sibling one import further in, see below), moved to eeschema/ under KiCad's own path where one exists.

dialogs/*.tsx → eeschema/dialogs/ (KiCad's own name, matching a real dialog_*.cpp): dialog_edit_symbols_libid, dialog_export_netlist, dialog_global_edit_text_and_graphics, dialog_image_properties, dialog_increment_annotations, dialog_label_properties, dialog_line_properties, dialog_sch_import_settings, dialog_schematic_setup, dialog_shape_properties, dialog_sheet_pin_properties, dialog_sheet_properties, dialog_symbol_fields_table, dialog_sync_sheet_pins, dialog_table_properties, dialog_tablecell_properties, dialog_text_properties, dialog_update_from_pcb, and the six dialogs/panels/panel_*.tsx (flattened into dialogs/, no panels/ subdirectory — KiCad's own dialogs/ has none either): panel_bom_presets, panel_eeschema_annotation_options, panel_setup_buses, panel_setup_formatting, panel_setup_pinmap, panel_setup_severities, panel_template_fieldnames. dialog_import_gfx.tsx → eeschema/import_gfx/dialog_import_gfx_sch.tsx (KiCad's actual path: the schematic importer, as opposed to the footprint-editor one). ErcDialog.tsx → eeschema/dialogs/dialog_erc.tsx (DIALOG_ERC).

Widgets with a KiCad name: SearchPanel.tsx → eeschema/widgets/sch_search_pane.tsx (SCH_SEARCH_PANE, widgets/sch_search_pane.cpp).

Kept our own name, moved to the eeschema root anyway (no single KiCad translation unit, same as pcbnew/STRUCTURE.md's stage-A leftovers table): item_color.ts (colour-control glue every moved dialog needs — dialogs/), NetNavigatorPanel.tsx → eeschema/widgets/net_navigator_panel.tsx (net_navigator.cpp's UI half is inline in SCH_EDIT_FRAME, no standalone widget file to match), pdf_annotations.ts, pin_alt_icon.ts, render_color.ts, symbol_markers.ts (all three fragments of sch_painter.cpp, which has no whole .ts port yet to land beside — flat at the eeschema root, since KiCad's eeschema/ has no render/ directory), global_sym_lib_table.ts, lib_tree_item.ts, project_sym_lib_table.ts, repair_source.ts (same reasoning, no symbols/ directory in KiCad's tree either). project_settings.ts moves too, for the same reason.

panel_setup_severities.tsx duplicates common/dialogs/panel_setup_severities.tsx (the real shared panel) — moved alongside dialog_schematic_setup.tsx because it had to (severing that relative import was the only alternative), but the duplication itself is not fixed here, same as pcbnew/STRUCTURE.md's identical finding for its own panel_pcb_severities.tsx.

Not moved, transitively designer/-coupled (surfaced by tsc, not by the first relative-import scan, which only checked one hop):

  • dialog_change_symbols.tsx and symbol_chooser_frame.tsx — both reach widgets/panel_symbol_chooser.tsx, which needs editors/pcb/footprint_preview_panel.tsx / widgets/footprint_list.ts / prefs/settings.ts. Same shape as pcbnew/STRUCTURE.md's footprint_chooser_frame.tsx finding.
  • render/plot.ts (SCH_PLOTTER/SCH_PRINTOUT) — needs theme.ts and render/renderer.ts, both genuinely designer/-coupled (renderer.ts needs ui/view_controls.ts, font/outline_fonts.ts, font/draw_outline_text.ts). Same "not attempted this pass" verdict pcbnew/STRUCTURE.md gives pcbTheme.ts/renderBoard.ts.
  • symbols/preload_pool.ts — needs symbols/preload_worker.ts, which needs libraryBundleStore.ts (real cloud/library plumbing).

eeschema/tsconfig.json gained jsx: "react-jsx" and DOM/DOM.Iterable here, the same shape pcbnew/tsconfig.json picked up for its own first .tsx move; package.json gained react, @types/react, vite and fflate (the last for dialog_export_netlist.tsx's zip export).

What's still in designer/, and why stage E2 pt 2 (EESCHEMA_APP) has no precedent to copy yet

SchematicEditor.tsx itself (11.5k lines) and SymbolEditor.tsx (2.9k lines) — the windows — are unmoved. What is left of SchematicEditor.tsx's app plumbing after pt 1 is the same shape pcbnew/STRUCTURE.md lists for PcbEditor.tsx: PreferencesDialog, HomeLink, OpenFileDialog/ SaveAsDialog, useToolbarEntries, PresencePanel, useAuth (AuthProvider), useProjectSync, settings/gridSizeToIU (prefs/settings.ts), DialogEeschemaPageSettings, fetchNetlistFromSchematic (editors/pcb/netlist_from_schematic.ts) — all real app-level plumbing an EESCHEMA_APP interface would need to abstract, the same way CVPCB_APP (cvpcb/cvpcb_mainframe_ui.tsx) already does for cvpcb/.

Worth flagging before building one: pcbnew/STRUCTURE.md's own "Not moved, and why" section says PcbEditor.tsx is also still unmoved, waiting on a PCBNEW_APP that "none of which is written yet" — the pattern this task was asked to copy has not itself finished the app-interface half for its own, larger frame. CVPCB_APP is the only complete worked example in the tree (small program, three windows). An EESCHEMA_APP for an 11.5k-line frame is a bigger interface than either precedent and was not attempted in this pass; the dependency list above is the resume point.

The symbol editor (designer/src/editors/symbol/* → eeschema/symbol_editor/, SymbolEditor.tsx → eeschema/symbol_editor/symbol_edit_frame_ui.tsx) was not started this pass either, for the same budget reason.

Stage E2 pt 2, attempt (2026-09-28): EESCHEMA_APP not attempted, symbol

editor's six clean files moved

EESCHEMA_APP + the SchematicEditor.tsx move: not done, two reasons. First, the size gap from CVPCB_APP's precedent is real, not just larger — CVPCB_APP is 63 members covering three windows totalling ~1.7k lines; SchematicEditor.tsx alone is 11.5k lines with ~19 designer/-only imports reached from call sites scattered across the whole file (settings.*, useAuth(), <HomeLink/>, <PreferencesDialog/>, <SchematicCanvas/>, <FootprintChooserFrame/>, …), not concentrated the way FootprintCanvas/ Viewer3DFrame are cvpcb's only two JSX-returning interface members. Turning every one of those call sites into app.X is a refactor across thousands of lines, not a sed over import lines — the "moves and seams only, small commits" method this whole file otherwise uses does not cover it safely in one pass. Second, and decisive for this pass specifically: designer/src/editors/schematic/SchematicEditor.tsx had another agent's uncommitted changes in the working tree the whole time (a live, shared checkout — see CLAUDE.md's "Concurrent session branch hazard"), so touching it at all this pass would have been the one thing the commit rules categorically forbid, independent of the size question.

The task's second instruction — split sch_base_frame, sch_draw_panel, sch_view, sch_painter, sch_preview_panel, sch_plotter, sch_render_settings, eeschema_config, picksymbol, sheet out "where clearly separable" — turned up nothing to move: none of the ten exists as a file anywhere in the tree yet (checked by name), and inside SchematicEditor.tsx each is only a comment naming the C++ counterpart of code woven through the surrounding JSX (e.g. sch_render_settings/ sch_painter/eeschema_config are three or four sentences apiece inside applyRenderSettings-style blocks, not standalone functions). Pulling one out clean would be new module-boundary design, not a move — the render-path half of this (sch_painter/sch_view/sch_preview_panel) is the same render/renderer.ts+render/plot.ts+theme.ts cluster the previous pass already flagged "genuinely designer/-coupled ... not attempted this pass".

The exact designer/-only surface EESCHEMA_APP would need to cover, read off SchematicEditor.tsx's own imports (supersedes the prose list two sections up): auth/AuthProvider.js (useAuth), dialogs/ dialog_eeschema_page_settings.js (DialogEeschemaPageSettings), dialogs/ PreferencesDialog.js, dialogs/prefs/types.js (PrefsPageId), fs/ OpenFileDialog.js, fs/SaveAsDialog.js, prefs/settings.js (settings, gridSizeToIU), prefs/useSettings.js, sync/ProjectSyncProvider.js (useProjectSync), sync/ProjectSyncTransport.js, sync/sch_diff.js, ui/ HomeLink.js, ui/PresencePanel.js, ui/SelectionFilterPanel.js, ui/ useToolbarEntries.js, widgets/dialog_sch_find.js, widgets/ dialog_sym_lib_table.js, widgets/footprint_list.js, and one cross-editor one, ../pcb/dialogs/footprint_chooser_frame.js. Plus, from inside editors/schematic/ itself, the modules already known to be transitively designer/-coupled and therefore also EESCHEMA_APP candidates rather than plain moves: components/SchematicCanvas.tsx (own JSX member, the FootprintCanvas pattern), dialogs/dialog_change_symbols.tsx, render/ plot.ts, render/renderer.ts, symbols/preload_pool.ts, widgets/panel_symbol_chooser.tsx. This is the resume point; still nobody's built it, for either frame.

Symbol editor (task step 3): six of ~20 files moved, the rest blocked the same way. Read as a fresh check rather than trusting the prior pass's one-hop scan, every file in designer/src/editors/symbol/ was checked for designer/-only imports transitively (not just one relative-import hop — the same miss this file's own E2 pt 1 note warns about). Six had none, ever, at any hop, and moved: symbol_edit_frame.ts (unchanged name — SYMBOL_ EDIT_FRAME's KIWAY half, same shape as eeschema/sch_edit_frame.ts), frame_title.ts and delete_symbol_prompt.ts (kept under their own names, both fragments of symbol_editor.cpp, same "kept under our own name" reason as the fused-file table above), symbolToolbars.ts → symbol_editor/ toolbars_symbol_editor.ts (KiCad's own name), cursors.ts (own name, no KiCad file, same as pcbnew/footprint_cursors.ts's reasoning) — all four to eeschema/symbol_editor/, matching symbol_edit_frame.{cpp,h} etc. sitting under that directory upstream — and symbol_tree_synchronizing_ adapter.ts to the eeschema root, because KiCad's own symbol_tree_synchronizing_adapter.cpp lives at eeschema/, not eeschema/symbol_editor/.

Stage F (subfolders, 2026-09-29): frame_title.ts and delete_symbol_prompt.ts merged into symbol_editor.ts. Both were fragments split out of SYMBOL_EDIT_FRAME::UpdateTitle and SYMBOL_EDIT_FRAME::DeleteSymbolFromLibrary, and KiCad's own symbol_editor.cpp is the one file that holds both — so "kept under their own names" above is superseded; they now live at eeschema/symbol_editor/symbol_editor.ts (the two sections kept separate, each with its own doc comment). SymbolEditor.tsx and the two qa tests that imported the old paths were updated in the same commit.

The rest of the directory is more interconnected than the "no designer/ import" grep alone shows, three chains deep:

  • libraryManager.ts reaches designer/src/libraryHosts.ts directly (the same seam CVPCB_APP.footprintsBase() abstracts for the footprint editor's own libraryManager.ts — nothing wires the equivalent for symbols yet).
  • edits.ts reaches ./grid.js (symbolGridIU), and grid.ts reaches prefs/settings.js for SymbolEditorSettings/gridSizeToIU — a central store that stays in designer/ permanently by design (same reasoning as eeschema_settings.cpp in the "Not ported" table above), so edits.ts cannot move without a seam either. conditions.ts and menubar.ts both import edits.ts (directly or via conditions.ts), so the same block reaches them too.
  • render/symbolRenderer.ts and components/dialogs.tsx both reach editors/schematic/theme.ts — itself designer/-import-clean and a plausible eeschema-root move on its own, but moving it does not unblock either file, since both also reach the edits.ts→grid.ts chain above through other imports.

None of SymbolCanvas.tsx, SymbolEditor.tsx, grid.ts, defaults.ts, toggles.ts (same prefs/settings.js coupling as grid.ts), or prefs/*.tsx (the six Preferences-dialog panels — PrefsContext, ui/ action_catalogue.js, pcm/pcmStore.js, all central-dialog plumbing, the same category pcbnew/STRUCTURE.md and this file both leave for stage E2's app-interface work rather than move) went anywhere this pass. A SYMBOL_EDIT_FRAME_APP covering libraryHosts, prefs/settings's symbol slice, and SchematicCanvas-style theme access is the resume point for the rest of this directory — the same shape of gap EESCHEMA_APP has for the schematic editor above, one size class smaller.

qa/probes/struct_diff_eeschema.sh's root-name-match count after this stage: 48 of 66 eeschema-root .ts/.tsx files are 1:1 KiCad root-name matches (up from 45/28 before this stage; most of the delta is concurrent work from other agents on this branch, not this pass — this pass's own contribution is symbol_tree_synchronizing_adapter.ts, +1/+1). SAME in the regenerated probe output is 94 (up from 72).

Stage E3 pt 1 (2026-09-28): the live item classes, and the s-expression reader/writer

KiCad 10.0.5's schematic model built alongside the record model (types.ts and its EditCommands are untouched; no caller is switched yet), file for file:

  • Items: sch_item, sch_line, sch_junction, sch_no_connect, sch_bus_entry, sch_field (class appended), sch_text, sch_label (every label kind), sch_textbox, sch_shape, sch_bitmap, sch_table, sch_tablecell, sch_group, sch_rule_area, symbol, lib_symbol (class appended), sch_pin (class appended), sch_symbol (class appended beside the record getters), sch_sheet, sch_sheet_pin, sch_sheet_path (SCH_SHEET_PATH/LIST and the instance/variant records appended), sch_screen (+ SCH_SCREENS), sch_rtree (EE_RTREE), schematic (class appended), schematic_settings (SCHEMATIC_SETTINGS appended), bus_alias, default_values, sch_file_versions; junction_helpers gained the live AnalyzePoint.
  • sch_io/kicad_sexpr/: sch_io_kicad_sexpr_parser, sch_io_kicad_sexpr, sch_io_kicad_sexpr_lib_cache, sch_io_kicad_sexpr_common. A load takes a readFile( absolutePath ) callback (no disk); a save returns the text.

Deliberate divergences, each marked in its file:

  • EE_RTREE keeps items per type in insertion order and tests boxes when asked, not when inserted (a query sees where an item is now).
  • SCH_SYMBOL::UpdatePins reuses spare pins in m_pins order; upstream walks a std::set<SCH_PIN*>, i.e. allocation addresses.
  • SCH_SHEET_PATH's hash is the KIID path text.
  • SCHEMATIC::Settings() / ErcSettings() are schematic-owned (no live project file).
  • Pending: plotting, SCH_CONNECTION / CONNECTION_GRAPH, SIM models, SCH_COMMIT, PIN_LAYOUT_CACHE (so a pin's bounding box, and any spatial query that reaches one, throws), the fontconfig cache (fonts resolve by name, no embedded fonts), the symbol-library plugin half (library table), the clipboard Format.

Round-trip oracle (qa/data/eeschema/sexpr_oracle, kicad-cli 10.0.6 sch|sym upgrade --force run twice, as JobUpgrade loads and saves): KiCad's rewrite read and written by us is byte-identical for all 34 schematics and 6 symbol libraries; originals are byte-identical for 23 schematics and all libraries, and equal for the other 10 once KiCad's freshly invented uuids and its pointer-ordered pin runs are normalised.

Stage E3b (2026-09-28): markers, connections, the graph, commits on the live items

One KiCad class per commit, each onto the live SCH_* model (nothing in the app calls these yet — the record-model engine stays wired until a later stage switches the callers).

  • sch_marker.ts (SCH_MARKER). It needed ERC_ITEM, so erc/erc_item.ts came with it (the allItemTypes table and the per-item sheet paths), and erc/erc_settings.ts gained ERCE_T and a live ERC_SETTINGS class (the severity map and GetSeverity's pin-to-pin / duplicate-pin / generic special cases) beside the record model's ErcSettings. SCHEMATIC::ErcSettings() is schematic-owned, as Settings() already is — there is no live project file yet. Root match after: 54/83.

Stage E2 pt 3 (2026-09-28): the frame files out of SchematicEditor.tsx and render/

Moves only; every step is its own commit, importers repointed by qa/probes/relocate_imports.mjs (whole-module or named-export moves) and self-alias imports rewritten by qa/probes/deself_imports.mjs.

Step 1 — the render cluster (render/renderer.ts, render/plot.ts, theme.ts, which the previous pass called "genuinely designer/-coupled"): it was not. theme.ts only ever imported @ziroeda/common, and the one designer/ edge left in renderer.ts was drawField from the symbol editor's painter, which is itself pure. Order that made each step a plain move:

was now KiCad
editors/schematic/theme.ts + RenderOpts/Viewport/DEFAULT_RENDER_OPTS out of renderer.ts sch_render_settings.ts sch_render_settings.{h,cpp} (LoadColors + the m_Show* / ratio members)
editors/symbol/render/symbolRenderer.ts symbol_editor/symbol_renderer.ts (own name) the m_IsSymbolEditor half of sch_painter.cpp
editors/schematic/render/renderer.ts sch_painter.ts sch_painter.{h,cpp}
editors/schematic/render/plot.ts sch_plotter.ts sch_plotter.cpp + printing/sch_printout.cpp (still fused)

symbol_renderer.ts keeps our name: fusing it into sch_painter.ts as KiCad has it is not a move (both define MM, Viewport, grid helpers).

Two traps found on the way:

  • sch_painter.ts's '@ziroeda/eeschema' imports became './index.js', not the declaring modules — splitting them changes module evaluation order, and eeschema/'s index has cycles (project_settings.ts reads ERC_ITEMS from ./index.js at module scope).
  • central_values' walk of eeschema/ read .css/.tsx only, so moving renderer.ts dropped four colour sites from the ratchet (a test that cannot fail). It now reads .ts there too; pcbnew/ and 3d-viewer/ still do not (five uncounted .ts sites, listed in the test).

Root match after step 1: 58/83 (probe SAME 109; both include other sessions' concurrent SCH_* ports).

  • sch_connection.ts (SCH_CONNECTION, CONNECTION_TYPE), plus the per-sheet connection map on SCH_ITEM (Connection, InitializeConnection, GetOrInitConnection, SetConnectionGraph, GetEffectiveNetClass). SCH_ITEM creates its connections through a factory sch_connection.ts installs on load (SCH_ITEM.s_newConnection): a value import the other way would be a module cycle, since sch_connection.ts needs SCH_SHEET_PATH and sch_sheet_path.ts extends SCH_ITEM at load time. The graph is typed as the two calls the class makes of it (SCH_CONNECTION_GRAPH). The record model's IsBusLabel copies (junction_helpers.ts, sch_bus_entry.ts) are left for the caller switch. Root match after: 55/83.
  • connection_graph.ts (CONNECTION_SUBGRAPH, CONNECTION_GRAPH, NET_MAP), the whole of connection_graph.cpp including RunERC and its checks, on the live items. It is its own port rather than a wrapper over connectivity/ — the record-model engine is a union-find over plain records, a different algorithm from upstream's subgraph walk, so nothing there was the same code to reuse. Pinned by qa/unittests/eeschema/connection_graph_oracle.test.ts: the (nets …) section, written as makeListOfNets writes it, matches kicad-cli line for line on 45 designs — the 13 of qa/data/eeschema/netlist_oracle/ plus 32 more in netlist_oracle_graph/ (the rest of KiCad's netlists/ and the designs its own connectivity regressions load). Containers keyed by pointer upstream are insertion-ordered Map/Set here; the connection map is walked in std::less<VECTOR2I> order. Needed on the way: the live PIN_LAYOUT_CACHE bounding-box half (pin_layout_cache.ts, below the record model's copy; SCH_PIN::GetBoundingBox used to throw), and SCHEMATIC loading the project's bus aliases. SCH_SCREEN::GetLine asks the tree for lines only (same answer; EE_RTREE boxes every item per query, which made the 40-second video design take 3). Root match after: 62/83 (the root grew with other stages in between).
  • sch_commit.ts (SCH_COMMIT), over common/commit.ts' COMMIT. The frame, view and selection tool are reached through the tool manager as upstream does, each optional; what the commit calls on them is spelt out as three small interfaces (SCH_EDIT_FRAME_FOR_COMMIT, SYMBOL_EDIT_FRAME_FOR_COMMIT, SCH_SELECTION_TOOL_FOR_COMMIT), since neither frame is a live class yet. SCHEMATIC gained ConnectionGraph(), GetNetClassAssignmentCandidates() and RecalculateConnections() - always the whole-graph rebuild (upstream's incremental arm is an optimisation of the same answer; CleanUp is not on the live model yet) - and SCH_ITEM::Destroy now leaves the graph.
  • schematic_undo_redo.ts (SCH_EDIT_FRAME's SaveCopyInUndoList ×2, PutDataInPreviousState, RollbackSchematicFromUndo, ClearUndoORRedoList), as SCH_UNDO_REDO_MIXIN mixed into sch_edit_frame.ts's SCH_EDIT_FRAME — pcbnew/undo_redo.ts' pattern for one C++ class split over several .cpp files. The frame gained the live-model half those need (SetSchematic, GetScreen, GetCurrentSheet, AddToScreen/RemoveFromScreen, RecalculateConnections, the repeat-items list); the window still edits the record model and calls none of it. Not ported: the PAGESETTINGS command (DS_PROXY_UNDO_ITEM) and every view call (no live view). SCH_COMMIT's frame constructor takes an EDA_BASE_FRAME, since the live SCH_EDIT_FRAME is not a draw frame yet.

Stage F (subfolders, 2026-09-29): netlist_exporters/ closed, erc_report.cpp blocked

netlist_exporters/ is now 9/10 filename-matched (up from 3/10 at this stage's start): netlist.ts — the fused xml+orcadpcb2 file noted above — split into netlist_exporter_xml.ts (NETLIST_EXPORTER_XML), netlist_exporter_orcadpcb2.ts (NETLIST_EXPORTER_ORCADPCB2), and netlist_generator.ts (SCH_EDIT_FRAME::WriteNetListFile's counterpart file — the format → exporter dispatch). netlistCadstar/netlistPads, already written inside the old netlist.ts but never checked against kicad-cli, moved to their own netlist_exporter_cadstar.ts / netlist_exporter_pads.ts and picked up real fixes the oracle found: the *ADD_COM*/*PART* symbol list was reference-ordered and un-deduped where upstream sorts by uuid and skips a multi-unit reference's later units (findNextSymbol), and PADS was missing the blank line between *PART* and *NET*. Shared plumbing (symbolField, boardSymbols, the new sheetOrderedBoardSymbols, netPinsByName, NetlistMeta) now lives in a real netlist_exporter_base.ts (NETLIST_EXPORTER_BASE), which resolvePadNumbers also moved onto (out of sch_pin.ts — it is exporter infrastructure, not a SCH_PIN method). netlist_exporter_base.cpp's actual CreatePinList/eraseDuplicatePins/ findAllUnitsOfSymbol (the stacked-pin, user-net-over-auto-generated-net dedup that only OrcadPCB2::WriteNetlist calls) is still unported — it would change netlistOrcadPcb2's behavior, out of scope for a stage whose own rule was "stays byte-identical" for that file.

Byte-exact kicad-cli sch export netlist --format cadstar/pads oracle coverage: qa/unittests/eeschema/netlist_exporter_cadstar_oracle.test.ts / _pads_oracle.test.ts, over the single-sheet designs in qa/data/eeschema/netlist_oracle/ (both formats belong to the Export Netlist dialog, which — like netlist_exporter_xml/orcadpcb2 — exports the open sheet only; the CLI always exports the whole project, so a hierarchical design there is not comparable). test_multiunit_reannotate_2/_3 are excluded: kicad-cli gives a reference shared by two different, both blank-Value, library symbols a third symbol's value by some resolution this port does not reproduce — investigated, not explained, left open rather than guessed at. netlist.test.ts gained a synthetic multi-unit dedup regression test since none of the three oracle designs exercise one — a mutant of the dedup rule survived the oracle set but was caught by it.

netlist_exporters/ is now 10/10 filename-matched (2026-09-29, later same day): netlist_exporter_spice_model.cpp (NETLIST_EXPORTER_SPICE_MODEL) has its own file. Its four overrides of NETLIST_EXPORTER_SPICE (WriteHead/WriteTail/ReadSchematicAndLibraries/GenerateItemPinNetName) had already been folded into netlist_exporter_spice.ts's generateSpiceNetlist as its subcktName branch, in an earlier session — there is no exporter class here to subclass, so the override split naturally became one function's own conditional rather than a second file. That branch stays there (moving it here would duplicate the private readPorts/port-direction table the central-value rule forbids two copies of). What netlist_exporter_spice_model.ts does hold is what upstream's constructor and InstallPageSpiceModel give the page as its own identity: a named entry point, generateSpiceModelNetlist(sch, libById, aProjectName, netlist?, opts?), taking the project name rather than a raw options bag, and refusing the plain-SPICE .save/.probe checkboxes that NET_TYPE_SPICE_MODEL has no case for in createNetlist (dialog_export_netlist.cpp:523-526) — dialog_export_netlist.tsx's "SPICE Model" tab now calls it instead of generateSpiceNetlist with a raw subcktName. Byte-exact against kicad-cli sch export netlist --format spicemodel: qa/unittests/eeschema/netlist_exporter_spice_model.test.ts, three single-Device:R single-sheet designs (netlist_oracle/issue16003/untitled{,2}.kicad_sch, netlist_oracle/issue14657/issue14657_2.kicad_sch). Multi-item designs and bus-vector hierarchical-label ports are deliberately not in that fixture set: NETLIST_EXPORTER_SPICE::writeItems's item order for 2+ symbols (not declaration order) and vector-label expansion before readPorts sees them are both netlist_exporter_spice.ts's territory — confirmed by running kicad-cli --format spice on the same design and seeing the same order, i.e. a base-class quirk this file's four overrides do not touch — not something to fix from this file, which owns only the .subckt head/tail and the port-name substitution.

erc/erc_report.cpp (ERC_REPORT): investigated, not written. The pieces it needs are already ported and ready: RC_ITEM::ShowReport / GetJsonViolation (common/rc_item.ts), ERC_ITEM as a full RC_ITEM subclass with GetMainItemSheetPath/MainItemHasSheetPath (erc/erc_item.ts), and the JSON shape (RC_JSON.ERC_REPORT/ERC_SHEET, common/rc_json_schema.ts). pcbnew/drc/drc_report.ts is the exact model to follow once it is unblocked — same header/severity-counter/ignored-checks shape, ShowReport/GetJsonViolation called the same way.

The blocker: ShowReport/GetJsonViolation need real EDA_ITEM objects (an itemMap: Map<KIID, EDA_ITEM>, GetItemDescription() per item) — which only the live SCH_PIN/SCH_SYMBOL/etc. classes at the eeschema root provide (sch_pin.ts's SCH_PIN.GetItemDescription, for one). erc/erc.ts's runErc() runs against the plain-record Schematic/SchSymbol model (types.ts) instead, and returns a lightweight ErcViolation[] (code/severity/message/position/ref-id strings, not item references) — the same record-vs-live-class split already flagged for sch_io/sexpr vs sch_io/kicad_sexpr in the previous stage. runErc is also single-sheet only (erc_item.ts's own comment: "the ERC dialog is not on the live model yet"), where a faithful report needs the whole hierarchy. Writing a parallel, ad-hoc item-description generator inside erc_report.ts to dodge this would duplicate GetItemDescription against the central-value rule and drift from it; the concurrent eeschema live-model migration (this session alone landed connection_graph.ts, sch_commit.ts, schematic_undo_redo.ts, and — mid-session — a netlist_exporters/netlist_exporter_live_wip.ts porting NETLIST_EXPORTER_BASE/XML/KICAD onto the live SCHEMATIC — is already closing this gap from the other side. erc_report.ts is a small, mechanical port once ERC's engine (or a wrapper) hands it real item references.

Step 2 — EESCHEMA_APP and the window. eeschema/eeschema_app.ts states what the Schematic Editor's window asks of the program, member by member, each family its own commit; designer/src/editors/schematic/eeschema_app.tsx (useEeschemaApp() + SchematicEditorMount) is the only file that answers it. SchematicEditor.tsx then moved whole to eeschema/sch_edit_frame_ui.tsx and takes app as a prop (App.tsx mounts SchematicEditorMount, the PcbEditorMount shape). How each designer/-only reach was resolved:

reach resolution
prefs/settings + useSettings hooks app.settings (EESCHEMA_SETTINGS_STORE) + the four hooks; the eeschema.json shape moved to eeschema_settings.ts
Preferences, Open/Save As, HomeLink, selection filter, sym-lib table, footprint chooser, toolbars app members (component-shaped); PrefsPageId moved to common/frame_type.ts
hosted symbol/footprint libraries, preload, hotkey remap app members, stated structurally (no pcbnew type)
live sync, auth, presence sch_diff.ts + project_sync_transport.ts moved to eeschema/, collection_diff.ts to common/; hooks + panel are app members
CvPcb window, pcbnew netlist + cross-probe helpers app members (cvpcb imports eeschema: a direct import back is a package cycle). The cross-probe decision/flash belongs in common/; pcbnew/cross-probing.ts was busy
canvas contract moved to sch_draw_panel.ts; component is app.SchematicCanvas
symbol chooser, library browser, rescue, change symbols contracts moved (picksymbol.ts, project_rescue.ts, tools/change_symbols.ts); components are app members
annotate / print / plot dialogs moved to dialogs/dialog_annotate.tsx, printing/dialog_print.tsx, dialogs/dialog_plot_schematic.tsx, taking settings as a prop
pure helpers moved: hover_selection, net_overrides, panes, toggles, frame_title; TEMPLATES appended to common/template_fieldnames.ts; the field/symbol-properties/find/page-settings dialogs and the properties panel to dialogs/, widgets/

Every hook dependency list that reads an app member lists it (qa/probes/add_hook_dep.mjs); useExhaustiveDependencies stayed at 103 warnings throughout. Root match after step 2: 66/83 (probe SAME 134; both include other sessions' concurrent ports).

Step 3 — frame files out of sch_edit_frame_ui.tsx. Checked against each .cpp's own method list. What moved is code that is the C++ file's and reads none of the component's state:

KiCad file ours what
sch_draw_panel.{h,cpp} sch_draw_panel.ts the canvas contract (props, controller, pending placements) — step 2
picksymbol.cpp picksymbol.ts PICKED_SYMBOL and DIALOG_SYMBOL_CHOOSER's props/result — step 2
sheet.cpp sheet.ts InitSheet's new screen
eeschema_config.cpp eeschema_config.ts LoadProjectSettings' render-settings seed

Not moved, and why: sch_base_frame.cpp (GetLibSymbol/SchGetLibSymbol, SyncView, UpdateItem, HighlightSelectionFilter, RefreshZoomDependentItems) has no counterpart function in the frame — its behaviour is spread through React state and effects (and UpdateItem already lives on sch_edit_frame.ts/sch_commit.ts). sch_view.cpp and sch_preview_panel.cpp are the GAL view and the preview canvas; ours are designer's SchematicCanvas.tsx and widgets/symbol_preview_widget.tsx, reached through EESCHEMA_APP, not code inside the frame. picksymbol.cpp's SelectUnit / SelectBodyStyle / SetAltPinFunction and the rest of sheet.cpp (LoadSheetFromFile, CheckSheetForRecursion, EditSheetProperties) are closures over the component's state; pulling them out is a refactor, not a move. sheet_and_config.test.ts pins the two moved functions (neither was pinned before). Root match after step 3: see the regenerated struct_diff.

Step 4 — the Symbol Editor window. Same method as step 2:

was now
prefs/settings.ts's SymbolEditorSettings + defaults symbol_editor/symbol_editor_settings.ts, with setSymbolEditorSettingsProvider / currentSymbolEditorSettings (the live file, for the grid and item defaults)
gridSizeToIU (schematic IU) eeschema_settings.ts (common's two-argument one is the board editors')
editors/symbol/{grid,defaults,edits,conditions,toggles}.ts symbol_editor/ (own names)
editors/symbol/menubar.ts symbol_editor/menubar_symbol_editor.ts
editors/symbol/components/dialogs.tsx symbol_editor/symbol_editor_dialogs.tsx
ui/selection_filter_panel.ts + ui/SelectionFilterPanel.tsx widgets/panel_sch_selection_filter.ts + _ui.tsx
editors/symbol/libraryManager.ts symbol_library_manager.ts (symbols base is a constructor argument)
SymbolCanvas's contract sch_draw_panel.ts
editors/symbol/SymbolEditor.tsx symbol_editor/symbol_edit_frame_ui.tsx, taking SYMBOL_EDIT_FRAME_APP (symbol_editor/symbol_edit_frame_app.ts); designer's symbol_edit_frame_app.tsx builds it (SymbolEditorMount)

Still in designer/, each needing its own app seam: SymbolLibraryBrowser.tsx (SYMBOL_VIEWER_FRAME + toolbars_symbol_viewer), dialogs/symbol_chooser_frame.tsx and widgets/panel_symbol_chooser.tsx (SYMBOL_CHOOSER_FRAME / PANEL_SYMBOL_CHOOSER, with SYMBOL_TREE_MODEL_ADAPTER fused inside), and symbols/index.ts (the hosted library loader). Root match after step 4: 69/84 (the 10.0.6 tree has 84 root .cpps).

Own-named root files: the fold pass (2026-09-29)

Re-audit of the "fold audit" table above (2026-09-28), now that several of the files it declined for lack of a target have one. A KiCad root count here includes .h-only classes (bus_alias.h etc.), so it reads 95, not the step-4 count's 84 .cpp-only.

Folded (content moved bodily into the cited file, old file deleted, importers repointed with qa/probes/relocate_imports.mjs, one commit each):

ours folded into why now, not before
lib_symbol_compare.ts lib_symbol.ts Same target the audit already named; the "30+ importers, risk" objection was about doing it concurrently with other agents' edits, not about the merge itself — done with the private-index split-commit technique so it never touched a file mid-edit.
symbol_search_terms.ts lib_symbol.ts Same target, only 3 importers (not the 30+ of the file above) — the audit's own "considered and declined" note said the split was deliberate but never weighed the risk at this file's actual size.
render_color.ts, symbol_markers.ts, pin_alt_icon.ts sch_painter.ts The audit's reason for keeping them apart — "no eeschema/sch_painter.ts exists to fold into" — is no longer true; stage E2 pt 3 step 1 ported it. All three were already sch_painter.ts's own imports.
pdf_annotations.ts sch_plotter.ts Same shape: the audit's blocker ("no sch_plotter.ts exists either") is gone since the same stage.
panes.ts sch_edit_frame.ts Not in the original audit table (it was in the "kept in designer/" list). Cites two files (eeschema_settings.cpp's pane infos, sch_edit_frame.cpp's AddPane/updateSelectionFilterVisbility) as one dock model — schLeftDockLayout reads both tables together, so a split by data source would break the one thing the module states. Folded whole into the frame that owns the AUI manager, with the two-file citation spelled out in the doc comment instead of enforced as a directory split.
frame_title.ts sch_edit_frame.ts The audit called this "closest to moving" and declined only because splitting a class file further was thought too risky mid-task; it is a single whole method (updateTitle), no data-source split needed.

Moved to eeschema/browser/ (no KiCad counterpart at all — same reasoning eeschema_app.ts and sch_diff/repair_source already carried, now grouped the way pcbnew/browser/ groups pcbnew's own browser-only root files): project_sync_transport.ts, sch_diff.ts, repair_source.ts. eeschema_app.ts stayed at the root then — the frames helper's file, in active use throughout that stage; it joined them on 2026-10-10.

Still declined, re-checked, no change:

ours why it stays
project_settings.ts Still no single fold target — readSchematicSetupText/writeSchematicSetupText are ONE JSON round-trip pass apiece that reads/writes schematic.*, erc.*, net_settings.* and text_variables together in the same function, merging each back without clobbering keys it doesn't own. Splitting it by upstream file means splitting those two functions' control flow, not just moving text — a real restructure, not a move, and one that reaches into erc/erc_settings.ts (the connectivity agent's file) for a two-file split when the actual citation is four files. Declined for this stage on risk, same call the 2026-09-28 audit made, still correct under 10.0.6.
project_sym_lib_table.ts The audit's own finding holds: eeschema/symbol_lib_table.cpp does not exist anywhere in the 10.0.6 tree either (checked fresh, find-wide). SYMBOL_LIBRARY_ADAPTER (eeschema/libraries/symbol_library_adapter.{h,cpp}) is architecturally different — an async, sub-library-aware LIBRARY_MANAGER_ADAPTER, not a sym-lib-table sexpr parser — so renaming to match it would misrepresent the port, not fix the citation.
fieldbox.ts Genuinely multi-file (common/eda_text.cpp, common/gr_text.cpp, common/font/font.cpp, common/font/stroke_font.cpp, eeschema/sch_field.cpp, sch_painter.cpp all cited in its own doc comment) and has 18 importers across dialogs/, tools/, root and four qa/ oracle tests — the same shape as symbol_search_terms.ts before its fold, except no single target file exists for it the way lib_symbol.ts did.
net_overrides.ts Cites SCH_LINE/SCH_JUNCTION methods, both of which exist as live classes now, but the module itself is a whole-netlist pass (computeNetClassOverrides walks every net once) over both item kinds at once, not a per-item method — the same "one coherent computation over two upstream files" shape panes.ts was, except here neither sch_line.ts nor sch_junction.ts is the natural single owner the way sch_edit_frame.ts was for the dock. Only 1 importer, so low risk if a later pass wants to split the per-item math out; left alone this stage.
hover_selection.ts Cites SCH_SELECTION_TOOL::Main (sch_selection_tool.cpp) — that file has no port in eeschema/tools/ yet (checked: no match), so there is still nothing to fold into.
toggles.ts, global_sym_lib_table.ts, lib_tree_item.ts, project.ts Unchanged from the audit: no upstream file (toggles.ts, session-state UI bucket like the symbol/gerbview editors' own), a data file with no .cpp (global_sym_lib_table.ts), already at its one true match (lib_tree_item.ts — include/lib_tree_item.h, which is not under eeschema/ upstream either, so it was never really an "extra" file), or unrelated code sharing a word (project.ts vs common/project.cpp).
types.ts The old record-model engine. Still not dissolved — E3's live SCH_* classes exist alongside it (stage E3b), but nothing has switched callers over yet; forcing a merge here would be inventing a caller-switch this stage never attempted, not a move.

Root EXTRA count (this file's own .ts/.tsx root files with no KiCad root .cpp/.h match): 23 → 12. docs/eeschema-structure-diff.md regenerated; SAME 138 → 141 (the small residual delta between "23 fewer EXTRA" and "3 higher SAME" is other agents' concurrent commits on this branch during the same window, not this pass's own miscount).

Stage S1 (2026-10-03): the editor onto the live model

Plan and inventory: docs/eeschema-live-stage0.md. Structure follows KiCad 10.0.6 itself.

  • schematic_holder.ts — schematic_holder.h (SCHEMATIC_HOLDER); SCH_EDIT_FRAME implements it and registers in SetSchematic (sch_edit_frame.cpp:179).
  • schematic.ts gained CleanUp, RecomputeIntersheetRefs, ResolveERCExclusions, RecordERCExclusions, ResolveERCExclusionsPostUpdate, and RecalculateConnections takes upstream's arguments in upstream's order. Divergence: the graph is always rebuilt whole (upstream's GLOBAL_CLEANUP arm); no SCH_SELECTION_TOOL on the live model yet, so CleanUp's selection bookkeeping has nothing to update.
  • files-io.ts gained SCH_FILES_IO_MIXIN.OpenProjectFiles (files-io.cpp:98-790), mixed into SCH_EDIT_FRAME as schematic_undo_redo.ts is. Left to the window: lock file, save prompt, the create question (as KICTL_CREATE), progress, info bar, autosave, window state, saving the outgoing project's local settings, DIALOG_MIGRATE_BUSES. Not live yet: the legacy plugin, MigrateSimModels, LoadProjectSettings/LoadDrawingSheet.
  • sch_record_bridge.ts — transitional, no KiCad counterpart, deleted at S7. Opens the window's records through OpenProjectFiles (rebuilt on demand when a record changed), as pcbnew's commitViewToBoard bridged its view during #636.
  • sch_io/kicad_sexpr/sch_io_kicad_sexpr.ts exports its PosixPath (wxFileName on POSIX paths) for files-io.ts.

Stage S2 (2026-10-03): annotation core and the project's settings

  • sch_reference_list.ts gained the live SCH_REFERENCE / SCH_REFERENCE_LIST (sch_reference_list.cpp), beside the record functions until S7. Divergence: SortBySymbolPtr groups by first appearance (upstream orders by address, which only groups).
  • refdes_tracker.ts: the class is REFDES_TRACKER, as upstream. Its m_prefixData is a std::unordered_map, and Serialize writes it in that map's order, so it is held in common/libc/unordered_map.ts (libstdc++'s order, measured) rather than a sorted Map.
  • sch_sheet_path.ts gained GetSymbols, AppendSymbol, GetMultiUnitSymbols, AppendMultiUnitSymbol, GetSymbolsWithinPath and AnnotatePowerSymbols.
  • schematic_settings.ts / erc/erc_settings.ts: SCHEMATIC_SETTINGS and ERC_SETTINGS are NESTED_SETTINGS at schematic and erc of the PROJECT_FILE, made and loaded by SCHEMATIC::SetProject, and Settings() / ErcSettings() answer the project's (a schematic with no project file gets an unparented default, a case upstream does not have). Not here: NGSPICE_SETTINGS, and the EESCHEMA_SETTINGS app config the constructor takes its defaults from. The BOM parameters go through common/settings/bom_settings.ts' to_json / from_json ports.
  • erc/erc_item.ts: the item templates are built on first read. ERCE_T is in erc_settings.ts (erc_settings.h) and that module imports this one for rule_severities; built at load, a template would read ERCE_T before its module had run.