Files
vibemake-pcb-studio/docs/editor-status.md
T
likang 7e01147566
CI / typecheck + test + build (push) Has been cancelled
Initial commit: import VibeMake PCB Studio
2026-11-03 11:20:08 +08:00

28 KiB

Which editors are done, and what "done" was measured against

Ten launchers, each a port of a KiCad 10.0.5 frame. This is the record of how far each has been taken and, more importantly, what kind of evidence backs the claim. "Audited and matching" has already turned out not to mean "identical" more than once.

The four levels of evidence

Ranked by how much they are worth. A claim of completeness should say which of these it rests on.

evidence what it cannot catch
E1 code read against the pinned C++ anything the code does not say, such as a control that renders but is never reached
E2 engine tests with a real KiCad oracle (its QA vectors, or files it wrote) whether the panel reaches the engine at all
E3 the UI compared against a running KiCad, by a person or by measurement behaviour: a field that displays and is then discarded
E4 the frame driven end to end, output compared to real KiCad's nothing known. This is the standard.

The gap between E2 and E4 is not academic. The PCB Calculator passed 99/99 of KiCad's own QA vectors and a visual check, and five of its fourteen panels still gave wrong answers, because the vectors called the engine directly and nobody had typed into a box. Two of the tests that "covered" one of those bugs asserted zDiff === 2 * z0Odd, an expectation computed from the code under test, which passed while the variable was wrong.

Status

Complete

editor evidence closed by
PCB Calculator E1 + E2 + E3 + E4 #108, PRs #613 #617
Image Converter E1 + E2 + E3 + E4 #105, PR #612

PCB Calculator. About 180 displayed strings were read off real pcb_calculator over the accessibility bus, and 179 match character for character. The one that does not is an E24 resistor formula whose answer depends on unspecified std::sort tie-breaking, and KiCad's own ranking rule prefers ours. That is documented in the test with the derivation, not baselined. 106 settings keys are pinned per entry against a table extracted mechanically from pcb_calculator_settings.cpp. All 39 unit selectors were swept. The art files are byte-identical to KiCad's own SVG sources.

Image Converter. Conversion is compared against files the real bitmap2component wrote (qa/data/bitmap2component/), the threshold and scroller were driven when it was built, and the appearance was checked by Akshay. Its output is deliberately not byte-identical; see the memory bitmap2component-output-format. That decision is due a revisit now that the bare-atom reader gap is closed (PR #616).

A correction worth keeping. The Drawing Sheet Editor was listed here as complete after two audits had closed its frame and its tools. A third pass over the six source files neither audit had opened found fifteen further gaps (#619): invented status strings, failures shown in the status line where upstream raises a modal, one printed page where upstream prints two, and a Preferences dialog with two checkboxes against upstream's four pages.

Nothing about the earlier work was wrong. Those audits had a scope, the scope was not the whole editor, and "complete" was read as though it were. A completeness claim is only as wide as what was actually opened. That is why the table above names the kind of evidence, and why #619 names the files.

A fourth pass has now driven it end to end - see below. That is E4 evidence and it found more, including four things all three earlier read-only passes had looked straight past. It also settles the question the table is for: the editor is still not complete, and the reason is one named page group, not a fog.

A fifth pass closed three of that group's four pages, and named the fourth as an app-wide item rather than this editor's: no launcher here had Toolbars, because none had the TOOLBAR_SETTINGS store it edits, and building it for pl_editor alone would have been the per-launcher copy the Preferences dialog was split up to stop.

A sixth pass has since built it as app-wide work, which is the only way it was ever going to be built. The store went in first (designer/src/ui/toolbar_config.ts), then one PANEL_TOOLBAR_CUSTOMIZATION (designer/src/dialogs/prefs/), and the page now ships under all three headings this port has - Schematic Editor, PCB Editor and Drawing Sheet Editor - with ten toolbars across the three frames reading the store rather than their module constants. So the distinction this table exists to make held: the row was app-wide, it was fixed app-wide, and pl_editor got it for free along with two editors that were never the ones being audited.

A seventh pass has closed the two gaps that sat inside the three pages the fifth one shipped. The Colors page's theme now reaches the canvas, and the Grids page has all five of PANEL_GRID_SETTINGS' buttons with DIALOG_GRID_SETTINGS behind two of them. Both are described below, with what each cost.

The entry stays in Not complete all the same, and the reason has moved: what is left is the list under "Still open, and named" - the infobar palette, the stale status-bar templates, and the rest - rather than a Preferences page.

Not complete

editor state
Drawing Sheet Editor (pl_editor) E1 + E2 + E3 + E4 everywhere except one Preferences page. Frame, tools, menubar, status bar, print and all the file commands are closed and were compared against the running program (PRs #604 #607 #614 #618, plus the E4 pass below). Grid dot, axis skip, the status-bar field widths and the grid origin were all measured off a live pl_editor. Preferences is now the shared dialog with three of upstream's four pages under it - Display Options, Grids and Colors, two of them the shared common/ panels rather than copies. Toolbars has since landed, as the app-wide work it always was: a TOOLBAR_SETTINGS store, one shared PANEL_TOOLBAR_CUSTOMIZATION, and the page under all three headings this port ships - so all four of upstream's pl_editor pages are now in. The two items inside the pages that did ship are closed too: the canvas reads GetColorSettings() the way PL_DRAW_PANEL_GAL's constructor loads it, and the Grids page has all five buttons plus DIALOG_GRID_SETTINGS. So all four Preferences pages are in and none of them is a control nothing reads. #619 is closed apart from the items below, which are chrome and measurement rather than pages.
Symbol Editor audited, PR #606. The enable/disable rules are now closed: all 53 setupUIConditions registrations ported per entry (PR #620) and the four the first mutation sweep could not tell apart pinned (PR #622). Open: LIB_TREE chrome, three missing dialogs.
Footprint Editor audited, PR #608. Open: seven items, headed by the dialog wall. dialog_pad_properties.cpp alone is 2492 lines against our 297 total.
GerbView exporter is a real port of GBR_TO_PCB_EXPORTER (PR #605). No mapping dialog, and aperture-macro holes export solid.
Schematic Editor tracker #195
PCB Editor tracker #200
3D Viewer, Project Manager not audited as units

Driving the Drawing Sheet Editor end to end

The first E4 pass over pl_editor. qa/probes/pl_e2e launches the installed /usr/bin/pl_editor on a private profile, clicks its menus over AT-SPI, types into its dialogs over XTEST and photographs its status bar, and every file command was run through it: New, Open, Open Recent, Append, Save, Save As, Print, Preferences, the page selector, a file that fails to parse and a file in an out-of-date format.

It ran twice, on two profiles. The second was launched with corner_origin = 1 and a 5 mm grid specifically to answer the grid-origin question below, and it also read the whole menubar back and photographed the Preferences tree. Close a driven pl_editor by pid; xkill once took out this desktop's mutter-x11-frames.

What that found, on top of #619's list:

  • Every status string was invented, and there were more of them than #619 counted. pl_editor writes the message pane in exactly five places, all in files.cpp, and the sentences are File '%s' saved. (after an Open, which is upstream's own slip, and after both saves), File '%s' inserted, File '%s' loaded, and - for New - nothing at all. It names the full path. Ours wrote its own sentence for each of those and for seven more events upstream leaves the pane alone for: placing an item, copying, resizing, pasting, the page dialog, a failed load and About.

  • Save added a row to Open Recent. UpdateFileHistory is called by LoadDrawingSheetFile and by nothing else; a driven pl_editor's Open Recent after a Save As does not list the file it just wrote. Ours listed every save.

  • The status bar is not the shared one. PL_EDITOR_FRAME builds its own dims[] and calls SetFieldsCount with it after the base constructor has set the shared widths (pl_editor_frame.cpp:150-181). Five of the eight panes differ, and one of them matters: pane 5 carries coord origin: Right Bottom page corner and is sized for it, where the shared table sizes pane 5 for the word Inches. Measured off the captured bar as well as read: field starts at 773, 858, 1034, 1224, 1340 and 1628 px on an 1854 px bar.

  • Append does not update the title. It pokes GetScreen()->SetContentModified() rather than going through OnModify() (files.cpp:150), so the * does not appear until the next real edit. A driven pl_editor shows probe - Drawing Sheet Editor straight after an Append, not *probe.

  • The page selector does not number the title block. DS_DRAW_ITEM_LIST starts at page "1" of 1 (ds_draw_item.h:409-410) and PL_DRAW_PANEL_GAL::DisplayDrawingSheet sets only the paper format, the title block and the project on its dummy list (pl_draw_panel_gal.cpp:100-103), so OnSelectPage toggles LAYER_DRAWINGSHEET_PAGE1 and _PAGEn visibility and leaves ${#} and ${##} alone (pl_editor_frame.cpp:461-467). A driven pl_editor reads Id: 1/1 on Page 1 and on Other pages, while 6266 canvas pixels change between the two - so the layers did toggle and the numbering did not. Ours read 2/2 on the second. The printout is the one place a number moves and only the numerator moves: PrintDrawingSheet is handed aScreen->GetPageCount(), which is BASE_SCREEN's 1 because pl_editor never calls SetPageCount, so the second sheet prints as page 2 of 1 (eda_draw_frame.cpp:1236-1239, base_screen.cpp:39,70-80). This one was found by driving the program, not by reading it: four passes over the C++ had gone past it.

  • dx -0 dy -0, and the shared %g was dropping the sign. #619's G11 said the X, Y - placeholder had no upstream equivalent, which is right; what it could not say is what stands there instead. UpdateStatusBar multiplies each axis by the origin corner's sign, so a pl_editor on Right Bottom page corner with A3 paper and no pointer over its canvas reads X 410 Y 287 - (0, 0) through that transform, 410 = 420 - 10 - and dx -0 dy -0, minus zeros included, because %.4g of 0 * -1 is -0 in C. Both photographed. Ours printed dx 0 dy 0: formatG in common/string_utils.ts returned '0' for any zero. That is a shared function every %g in the app goes through, so it was fixed there rather than worked around in the status bar.

Three items in #619 were misstated and are corrected rather than implemented:

  • The print preview is dead code. ToPrinter( true ) is the only path to wxPreviewFrame and SetZoom( 70 ), and PL_EDITOR_CONTROL::Print passes false (tools/pl_editor_control.cpp:116). Nothing else calls it, so there is no preview command to be missing - the same finding as DIALOG_NEW_DATAITEM.
  • The status-bar pane comment was backwards in the direction #619 said, but the fault was larger than a comment: the widths were wrong too.
  • RollbackFromUndo's non-PLUS branch calls Refresh() only, not HardRedraw(). Our restore path re-derives the properties panel and the message panel from state on every render, so the re-sync the PLUS branch does explicitly happens here by construction. No change was needed.

Two of #619's "unaudited, report as unknown" items, settled

#619 ended with a list of things it had not opened and asked for them to be reported as unknown rather than clean. Two of them are answers, not gaps:

  • The grid origin does not follow the origin selector, and it does not follow the setting either. PL_EDITOR_FRAME's constructor calls SetGridOrigin( ReturnCoordOriginCorner() ) (pl_editor_frame.cpp:218-219) and OnSelectCoordOriginCorner pointedly does not repeat it (:470-476), so #619 read this as ours possibly re-anchoring the grid on every render. It is the other way round and both are (0, 0): at that point in the constructor DS_DATA_MODEL::SetupDrawEnvironment has not run, so m_RB_Corner and m_LT_Corner are still zero and ReturnCoordOriginCorner() returns the origin whatever corner_origin says. Measured: a pl_editor launched on a profile with corner_origin = 1 and a 5 mm grid draws its lattice through the paper's top-left corner, 2.0 mm clear of the coordinate-origin marker it puts at the page's right-bottom - 287 mm mod 5 mm, exactly. Our grid is anchored at (0, 0) and originIU reaches the marker rather than the lattice, which is the same picture. ACTIONS::gridResetOrigin therefore has nothing to move in either program.
  • GetPageNumberOption() is dead code. It is declared, defined, and called from nowhere in 10.0.5 - the third such find in this editor after ToPrinter( true ) and DIALOG_NEW_DATAITEM. There is nothing to port.

Decisions taken rather than omissions

  • The temp file (SaveDrawingSheetFile writes to wxFileName::CreateTempFileName, copies permissions, renames over the target) buys the property that a serialiser which throws cannot truncate the file that was already there. serializeDrawingSheet returns the whole string or throws before anything reaches the store, so the call order gives the same guarantee and a rename would add nothing. DuplicatePermissions has no counterpart - the project store has no file modes. Worth knowing: upstream's own rename loses the mode on a new file, and a driven Save As really does leave 0600 behind.
  • Layout file is read only. is the other half of the load infobar and has no browser analogue; the outdated-format half is ported.
  • The print-failure sentence (An error occurred attempting to print the drawing sheet.) reports a printer that refused the job. A blocked popup is a different event, so ours says so in its own words rather than pointing the user at a printer.
  • Could not load image from '%s'. is reproduced with the %s, because wxMessageBox's second argument is the caption and upstream is passing the file name there. That is what a user sees.
  • The image chooser is the browser's. AddDrawingSheetItem's bitmap arm opens a wxFileDialog captioned Choose Image, starting in m_mruImagePath and remembering it afterwards (pl_editor_frame.cpp:863-871). Place > Image here is an <input type="file">, which the browser titles and positions itself; neither the caption nor a most-recently-used directory is reachable from a page. The failure message on the far side of it is ported.

Still open, and named

  • Preferences: all four pages are in. Three landed in the fifth pass; the fourth, Toolbars, landed in the sixth as app-wide work - see below. The dialog was opened on a running pl_editor and photographed: under Drawing Sheet Editor the tree carries exactly four pages - Display Options (PANEL_GAL_OPTIONS), Grids (PANEL_GRID_SETTINGS), Colors, Toolbars (PANEL_TOOLBAR_CUSTOMIZATION), registered at pagelayout_editor/pl_editor.cpp:68, 71, 82, 85.

    The local modal is gone. pl_editor now opens the shared PreferencesDialog every other launcher opens, and Display Options, Grids and Colors are pages of it. Two of the three are shared code rather than new files, which is the point: PANEL_PL_EDITOR_DISPLAY_OPTIONS is upstream nothing but an embedded PANEL_GAL_OPTIONS, and the Grids page is PANEL_GRID_SETTINGS

    • one class every KIFACE constructs with its own FRAME_T, which is the whole difference between one editor's page and another's. Both now exist once in designer/src/dialogs/prefs/, and eeschema's Display Options and Grids pages are the same two components. ACTIONS::gridProperties lands on the Grids page, which is the whole of what that action is.

    Folding eeschema's private copy in found that it had drifted from KiCad in four places nobody had looked at: a wxChoice where upstream has radio buttons, 45 degree crosshairs spelled 45 full window crosshairs, and both numeric ranges invented - grid thickness 1..5 against upstream's 0.5..10.0 by 0.5, minimum grid spacing 2..50 against 5..200 by 5. Those are fixed for both editors at once, which is what having one copy is for.

    Toolbars was declared absent rather than left to be noticed, and that is how it got built. OMITTED_PAGES in designer/src/dialogs/prefs/registry.ts named it with its reason, next to UPSTREAM_BOOK - upstream's own page list per heading, transcribed from EDA_BASE_FRAME::ShowPreferences - and a test required that shipped + declared-absent equal upstream's list, in upstream's order, per heading. The declaration is what made the gap addressable instead of invisible.

    It now ships, app-wide. Nothing about PANEL_TOOLBAR_CUSTOMIZATION is browser-hostile; the reason it was absent is that it edits a TOOLBAR_SETTINGS file this port did not have, every launcher's toolbars being module constants (editors/*/…Toolbars.ts), so the page would have had nothing to write to. That was fixed in the order the dependency runs:

    • designer/src/ui/toolbar_config.ts is the store, keyed as KiCad keys it - a toolbars list of { name, contents }, TOOLBAR_ITEM_TYPE and TOOLBAR_LOC spelled as magic_enum spells them - loaded free-form, because a stored toolbar replaces its default and merging the two would produce a toolbar neither side asked for. Each editor's existing …Toolbars.ts gained a DefaultToolbarConfig map, so no toolbar is transcribed twice.
    • common/dialogs/panel_toolbar_customization.tsx is the page, once, the way PANEL_GRID_SETTINGS is - three ten-line wrappers pass their app's settings, store and defaults, which is all the seven KIFACEs pass upstream.
    • the frames read it. EDA_BASE_FRAME::RecreateToolbars asks GetToolbarConfig( loc, m_CustomToolbars ) and never touches DefaultToolbarConfig; ui/useToolbarEntries.ts is that call, and all ten toolbars across the drawing sheet, schematic and board editors go through it. This is the half a customisation page is worthless without, so it is pinned twice: a rendered test that drives a real <Toolbar> and asserts on the buttons in the DOM, and a per-occurrence call-site test - one bar left wired to its constant is exactly the bug, and it would survive a check that the editor merely mentions the hook somewhere.

    Toolbars therefore moved out of OMITTED_PAGES under all three headings this port ships, and the page-book test gained an arm that requires every heading whose upstream list has a Toolbars row to ship one. Symbol Editor, Footprint Editor, 3D Viewer and Gerber Viewer have Toolbars pages upstream and no Preferences heading here at all, so theirs arrive with those headings.

    Two divergences are deliberate and stated at the seam. Upstream's TransferDataFromWindow writes all four toolbars back on OK, changed or not, so merely opening the page and pressing OK freezes that app's toolbars at today's defaults for ever; ours stores a toolbar only once it is edited. And ResetPanel empties the store instead of refilling it with the defaults - same drawn result, since GetToolbarConfig falls through to DefaultToolbarConfig when nothing is stored. Neither is visible on screen; a <app>-toolbars.json written by the two would differ.

    What is genuinely reduced, and is said on the page's own module rather than only here: upstream's action list is ACTION_MANAGER's whole registry filtered by isActionSupported, and ours is the union of that app's default toolbars. A button's icon and tooltip live on the ToolButton literal in the editor's toolbar module, so an action on no default toolbar has no literal to take them from. Closing that needs an action registry, not a bigger page.

    The same mechanism caught two things nobody was looking for. Our Schematic Editor heading carries an Annotation Options page upstream does not have there at all - PANEL_EESCHEMA_ANNOTATION_OPTIONS is a page of DIALOG_SCHEMATIC_SETUP, not of Preferences - now declared in EXTRA_PAGES and tracked with #195. And the PCB Editor heading ships one of its seven pages, which is #200's, but is now stated rather than implied.

    What was already closed stays closed: the invented "black background" checkbox is gone (the capture confirms pl_editor has no such control anywhere) and does not come back with the new pages - the test now checks all four pl_editor sources for it, not just the editor - and the crosshair is the three-way radio, Small crosshairs / Full window crosshairs / 45 degree crosshairs, plus a separate Always show crosshairs.

    The two things inside those pages are closed. Both were pl_editor's own rather than app-wide, and both were recorded before they were fixed:

    • The Colors page changes what is drawn. It stored appearance.color_theme and nothing read it. The colours enter in PL_DRAW_PANEL_GAL's own constructor - m_painter->GetSettings()->LoadColors( ::GetColorSettings( cfg->m_ColorTheme ) ) (pl_draw_panel_gal.cpp:57-59) - and DS_RENDER_SETTINGS::LoadColors takes exactly three layers of it (ds_painter.cpp:66-68): background, page border and sheet ink. The canvas calls that as a hook, because upstream it is the DRAW PANEL that asks and not the frame.

      It also settled black_background, which our canvas had been painting from. Upstream that never reaches the GAL: it sets m_drawBgColor (pl_editor_frame.cpp:541) while onPaint clears to settings->GetBackgroundColor() (draw_panel_gal.cpp:364). Its readers are the printer, the properties frame's swatch and DIALOG_PAGES_SETTINGS' preview, and it now goes to that preview.

    • The Grids page has all five buttons. Add, Edit, Move Up, Move Down and Remove (panel_grid_settings_base.cpp:34-56), each acting on m_currentGridCtrl's selection - which is a wxListBox whose selection is also the current grid (:194), so the radio-per-row and the per-row Remove went with it. DIALOG_GRID_SETTINGS is common/dialogs/dialog_grid_settings.tsx, in dialogs/ rather than dialogs/prefs/ because upstream it is common/dialogs/ and owned by no app. Grid size '%s' already exists. and Grid size X out of range. are both in.

      The stored shape had to change with it: window.grid.sizes held one string per grid where GRID_SETTINGS::grids holds GRID{ name, x, y } (grid_settings.h:33-54), so there was nowhere to put a name or a non-square Y. Stored files upgrade at load, and every row now renders through RebuildGridSizes' _( "%s%s (%s)" ), which is also what gives eeschema's View > Grid menu its second unit.

    Two bugs fell out of reading that C++ rather than watching the behaviour: a unitless grid string is millimetres and not mils (GRID::ToDouble passes EDA_UNITS::MM, grid_settings.cpp:53-57), and the dialog reads GetDoubleValue() rather than GetValue() (dialog_grid_settings.cpp:80), so a typed grid is stored unquantised.

    PANEL_GRID_SETTINGS is shared, and one launcher is still missing. Upstream pcbnew and the footprint editor mount it too, and OVERRIDE_ROWS already carries their rows; we have sch-grids and ds-grids only, because PcbnewSettings has no window.grid for the page to write. That is a pcbnew gap rather than a pl_editor one, and it is the next thing this panel wants.

    The grid list is now a setting rather than a read of DefaultGridSizeList(), because a page that edits a list nothing reads is the failure this whole audit exists to catch. The canvas and the grid context menu both index window.grid.sizes, which PL_EDITOR_DEFAULTS seeds from the shared table.

  • The infobar's palette. The strip uses the shared .ze-infobar the schematic raises. Its colours were ported for that editor and have not been measured against a live pl_editor's wxInfoBar, which is a dark bar with a red round icon rather than the amber one we draw.

  • The shared status-bar templates are out of date. STATUS_FIELD_TEMPLATES says X 00000.0000 Y 00000.0000 where 10.0.5 says X 1234.1234 Y 1234.1234 (eda_draw_frame.cpp:809-819), and the same for the delta and grid panes. That is every draw frame except pl_editor, which now states its own, so it is an app-wide item and not this editor's.

  • One character in one menu label. The whole menubar was read back over AT-SPI and compared item by item against ours: File, Edit, View, Place, Inspect, Preferences and Help match in order and in wording, Append and Reset Grid Origin included, and Append really is under Place rather than File. The single difference is that the running program's fifth File item is Save As… with a one-character ellipsis, where ACTIONS::saveAs' FriendlyName is Save As... with three dots (common/tool/actions.cpp:102) and ours follows the source. The literal is not in the installed libkicommon, so it is not settled whether the program, wx, GTK or the accessibility layer produced it, and the three-dot form is repo-wide here (qa/unittests/designer/menu_ellipsis.test.ts). Recorded rather than changed: one ambiguous observation is not enough to move a rule every launcher follows.

Library loading is an application-wide behaviour, not an editor's

Upstream does not load libraries when the place tool is invoked. IFACE::PreloadLibraries is scheduled by CallAfter the moment a project opens (sch_edit_frame.cpp:1492-1499, files-io.cpp:857-864, pcbnew/files.cpp:605-612, kicad_manager_frame.cpp:539-549), so the chooser opens on data that is already resident. The progress lives in the status bar and its BACKGROUND_JOB_LIST window, never in a dialog, and the chooser itself has no loading state at all.

Ported in PR #621. Two things are worth not re-deriving:

  • What the user was waiting on was measured, not guessed. Expanding one library in the chooser fetched every symbol in it as its own file: "Device" cost 536 requests and 2,486,139 B where the library file is 2,414,640 B, the same bytes to within 3 percent and 10 to 50 times the wall clock. Expand All did that for all 223 libraries, so one click asked for 22,778 files and 219.7 MB.
  • What is preloaded is the name index plus every symbol the open design places and every footprint it assigns, not the whole catalogue. That is precisely the set upstream's chooser reads synchronously in PANEL_SYMBOL_CHOOSER's constructor, so it is bounded by the design rather than by the catalogue. 219.7 MB in a browser tab would be a worse experience, not a more faithful one.

Things measured once, worth not re-deriving

  • Dialog geometry is remembered per dialog title. DIALOG_SHIM::Show restores a saved rect keyed by the title (dialog_shim.cpp:452-474), and eeschema and pcbnew share "Page Settings". Measured on a clean profile, pcbnew's true first open is 736 px wide against a used session's 923. We persist no dialog geometry at all, which is an app-wide gap rather than a Page Settings one. qa/probes/page_settings_width/ measures this.
  • KiCad is single-instance. Launching a second pcbnew opens a window in the running process, so a clean-profile measurement needs its own HOME.
  • The snap environment breaks a KiCad launch (__libc_pthread_init). Use env -i, as qa/probes/ does. A fresh profile shows a KiCad Setup dialog first.
  • The X root window cannot be captured on this Xwayland session (import -window root fails), but an individual toplevel can. Find the window id with xwininfo -root -tree first.