CHANGELOG.md

Changelog

All notable changes to DICOMKit will be documented in this file.

The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.

[Unreleased]

Added — linked film cells: adjust one, adjust them all (2026-08-07)

Work in progress, not yet committed.

"Apply this window to all cells" was a one-shot copy made after the fact. Judging a film means comparing its cells, and cells can only be compared when they are shown alike — which meant repeating every zoom by hand, cell by cell.

• The tools are back on a rail, and the locks are on it too. Window/level, zoom, pan, text, arrow and invert (V) sit down the left of the film with the three locks, the scope and the two resets. A right-click menu is a fine place for a command and a poor place for a mode: which tool is armed and which cells are linked have to be answerable at a glance, and a menu that only exists while it is open cannot answer them. The menu keeps everything it had.
• The lock is drawn on the cells it applies to. A closed padlock appears on every cell that will move when the focused cell is dragged — and on none that will not, which is how the scope becomes visible before the drag rather than after it. It is the one thing the preview draws over a picture that the film will not carry, sits in the corner over the letterbox margin, and takes no clicks.
• A mode, not a command. PrintCellSyncOptions — Zoom & Pan, Window, Invert — are locks on that rail. While one is on, dragging a cell carries that adjustment to the others as the gesture happens. Zoom and pan are one switch deliberately: cells that magnify together but sit over different anatomy are the confusing state, not one anybody asks for. Rotation and flip are not offered: they are how one image is put the right way up, and turning the whole sheet because one cell was upside down is never what was meant.
• A way back out. Reset Cell and Reset All sit beside the links (0 and ⇧0), because one linked drag can now put a whole sheet wrong and undoing it must not be harder than causing it. The cell menu's reset still acts on the cell that was right-clicked, and names itself so.
• A dragged cell keeps up with the drag. A window/level drag re-keys the cell on every mouse event, so its GPU render nearly always landed after its own key had been superseded — and PrintCellTextureCache threw those renders away, leaving the cell being dragged showing the picture the drag started from. A superseded render now stands in for the mark it belongs to, so the cell tracks the drag a dispatch behind instead of stalling.
• Geometry copies absolutely, windowing carries relatively. Every cell on a film is the same size, so the same zoom and pan is the same picture, and each peer's pan is then re-held inside its own image. A window is not portable that way — a film mixes modalities, and marks do not all state their window in the same space — so a drag scales each cell's width by the same factor and shifts its centre by the same fraction of a width. Presets are the exception: "lung" names a tissue, so it is copied as the numbers it is.
• Scope, because a film is not always one series. All Cells / Same Series / This Film, defaulting to Same Series — a film carrying two series is usually carrying them for comparison. Marks now carry their Series Instance UID for this; marks made without opening a file group by the folder they came from.
• It yields to the job. With raw pixels or a job-wide window on, nothing per-cell reaches the film, so the window link reads and behaves as off rather than claiming an effect the job has already taken away. Marks never opened are given their file's own window when the link goes on, so they move with the rest instead of sitting still.

Added — a film of one study names its patient once (2026-08-07)

Work in progress, not yet committed.

Identification was burned under every image, which is right for a sheet that mixes studies and repetitive noise on a sheet that does not: sixteen cells of one CT carried the same name sixteen times, each one costing its picture a strip of height. With multi-study selections coming, "print this film" has to answer both cases.

• The rule, in one place. FilmIdentificationPlanner decides per film, not per job: a sheet whose captioned images all share one Study Instance UID states the patient once along its bottom edge; a sheet mixing studies captions each image, as before. A job spilling onto a second sheet has each sheet decided on its own, because each sheet has to be identifiable on its own. The UID is the test rather than the caption text — two studies of one patient on the same day read identically and are still two studies. An image whose header could not be read has no study to agree with, so it forces per-image captions.
• The strip is kept clear of the pictures. FilmCellLayout takes a footer band out of the sheet before it lays out cells, so the footer sits under the bottom row rather than across it — text over anatomy is where a finding hides. Annotation boxes generally now get that band too; they used to be drawn into the bottom margin over whatever was there, and they are centred rather than flush left.
• It reaches real film three ways. Composed sheets (Save Film, the printer emulator) draw the footer themselves. On the wire it goes as a film-level Basic Annotation Box — PrintOptions.filmAnnotations carries a set per film, so two sheets of one job can name two patients, which one job-wide list could not.
• Whether the printer can carry it is asked, not assumed. The first cut looked at whether an Annotation Display Format ID had been typed into the settings sheet, so every printer that had not been configured by hand — which is every printer, by default — silently fell back to captioning each image and the film never matched the preview. Support is now a question about the association: DICOMPrintService.supportsAnnotationBoxes opens one, asks whether the Basic Annotation Box SOP Class was accepted, and releases. It is asked before the frames are prepared, because the answer decides whether they are captioned. A printer that takes annotation boxes but has no configured format ID gets a plain default (ANNOTATION), since a film box cannot carry annotation boxes without one; a printer that refuses the film box over that value has it created again without it rather than losing the job, and the progress line says the film will carry no annotation text. A printer that takes no annotation boxes at all — or cannot be reached — still gets the caption burned under each image: a film with no name on it is worse than a film that repeats one.
• The footer's type is sized off the film. It was a flat 3 mm, which is oversized on an 8×10 held in a hand and lost on a 14×17 across a viewing room. It is now 1.1% of the sheet's height, floored at 2.5 mm and capped at 6 mm — ~2.8 mm on 8×10 and ~4.75 mm on 14×17 — and the strip reserved for it, in the composer and in the preview alike, follows the type rather than a constant.
• Choosable, in the preview and in the settings column. Automatic (the rule above), "Under each image", or "Once at the foot of the film". The preview draws whichever the film will carry, scaled off the physical sheet, so the strip that is approved is the strip that prints.

Changed — the print preview stays shut, opens wider, and writes the film out (2026-08-07)

Work in progress, not yet committed.

• A launch shows the library and nothing else. Print Preview and Printer Emulator are singleton Window scenes, so macOS restored whichever was open at quit — the preview holding no marks and the emulator with its server stopped, because neither survives the app. Both now carry .restorationBehavior(.disabled) and .defaultLaunchBehavior(.suppressed); they open when they are asked for and not before.
• The console log column is as wide as the reader wants it. It was a fixed 300 points, which wrapped every import path and print job UID into four lines. It now opens at 460 and is dragged from the divider between it and the film, up to 60% of the panel and never below 260; the width is kept in AppStorage, so it is chosen once rather than every job. The film takes back whatever the log is not using, as it did before.
• The film can be saved as a file. "Save Film" in the preview's header writes PNG, TIFF or PDF. Not a screenshot of the preview: the images go through the same PrintService.prepare a real print sends them through and the sheet comes out of the same FilmComposer the printer emulator composes received film with, so what lands on disk is the sheet the printer would have laid down — identification band, drawn annotations, spillover and all. A PDF holds every film of the job as a page; PNG and TIFF write one file per sheet.

Added — film layouts the standard has and a grid has not (2026-08-06)

Work in progress, not yet committed.

PS3.3 C.13.3 lets a film's rows hold different numbers of images: ROW\1,3 is a scout above three slices, COL\1,2 is one image beside two. The Print SCP has always understood those forms — it has to, modalities send them — but everything that composes a film here could only say rows × columns, so the one layout a reader most often wants for a comparison film could be received and not sent.

• Seven band layouts are in the gallery, drawn. ROW\1,2, ROW\2,3, COL\1,2, COL\1,3, COL\1,4, COL\1,4,4 and COL\2,4,4 are picked the way the grids are — by looking at the film rather than reading a string — and live in PrintBandLayout in PrintOptionCatalog, the one table the print sheet and dicom-print --layout both read.
• The print sheet takes a format string too. Under the bands is a Custom field: type an Image Display Format and the film beside it is redrawn as it is typed, with the layout named in words underneath ("rows of 1, 3 — 4 images"). Picking a band fills that same field, so the layout in force can always be read as the string that will be sent, and adjusted from there. Text that is not a format says so in red and leaves the film on the automatic grid rather than quietly printing 1×1, which is what a lenient parse would have made of half-typed input. The gallery scrolls now that it holds four sections.
• The preview draws the film, not a grid. FilmPreviewView lays its cells out with FilmCellLayout — the same geometry the SCP composes received film with — instead of a SwiftUI Grid, so a band layout is shown as the bands it is. Every per-cell measurement (zoom and pan limits, annotation anchors, the identification strip) now comes from that cell's own rectangle; it used to come from one film-wide size, which was only ever right because the cells were all the same. Arrow-key navigation follows the cells' geometry for the same reason.
• The format reaches the printer verbatim. PrintLayoutSelection gained a .displayFormat case, PrintJobRequest/PrintPlan carry the format and count films by its image-box count, and the SCU sends the string as written. Verified end to end against dicom-printscp: --layout 'ROW\1,3' composes one image over three, 'COL\1,2' one beside two, 'COL\1,4,4' one beside two columns of four.
• dicom-print --layout accepts both. A grid token ("2x3") as before, or a format ('ROW\2,1,2', quoted so the shell keeps the backslash); the named bands are listed in its help. The dry-run banner and the film plan name a band layout by its format string — reporting ROW\1,1 as a 2×1 grid would have misstated the film.
• PrintImageDisplayFormat gained validated(_:) (strict, for UI and command lines), an initializer from a PrintLayout, isUniformGrid, and summary.

Changed — the viewer opens on the images (2026-08-06)

Work in progress, not yet committed.

Opening a study put four columns on screen before the first picture: the app's feature list, the series pane, the images, and an empty selection tray. Two of them were answering questions nobody had asked yet.

• The feature sidebar steps aside on the way into the viewer and comes back on the way out (MainView drives columnVisibility). The toolbar toggle still opens it over the images, and that choice holds until the viewer is left. Launching straight into the viewer starts collapsed too.
• The selection tray starts hidden (isPrintTrayVisible now defaults to false) and comes up on its own the moment the first image is marked — every marking path, the library's "Print…" included, goes through revealPrintTray(). Unmarking never puts it away again, so the "Clear" button cannot vanish under the pointer; opening a different study does, along with the marks it held.
• The current series is marked by one cue, on the card: a neutral white ring and a lifted surface. Nothing is drawn on the thumbnail — a ring there framed the letterboxing rather than the picture — and the accent stays out of the pane entirely, because in the viewer it means "this is what prints".
• In the grid, a marked tile no longer carries an accent edge along its bottom. The chip and the tick already say it, and with a film fully composed the edge drew a blue rule under every tile; the only accent edge left in the grid is the focus ring, so exactly one tile is lit — the one being worked on.

Changed — the viewer's three columns are told apart (2026-08-06)

Work in progress, not yet committed.

The series pane, the reading area and the selection tray all sat on the same near-black surface, so nothing on screen said which column held the images that were about to print:

• Three planes instead of one. The panes are a lighter surface (StudioColors.viewerPanel) with a dark seam on the side facing the picture; the gutter is darker than before (0.14 → 0.10); the reading area keeps its pure black and gains a shadow that lifts it off the mount.
• Every column is titled. A shared ViewerPaneHeader names the panes ("Series", "On film" with its count); the reading area's own strip names it, shows the tile layout, and reports "N of M on film" — the count of images on screen that are marked, so the middle column visibly owns the print selection.
• The reading area's frame is thicker and keeps the accent while it holds the keyboard, in a grid as well as at 1×1.
• Marked tiles are readable across a whole grid: a numbered film-position chip in the top-left corner (the number the tray lists it at) and an accent bar along the bottom edge. The border stays focus-only — the live tile's ring went 2 pt → 3 pt, and hovering brightens a tile's hairline.
• The patient plate in the series pane is neutral rather than accent-tinted: in the viewer the accent now means "this is what prints".

Changed — the print preview is a window, not a sheet (2026-08-06)

Work in progress, not yet committed.

• The preview's cells now draw from GPU textures (PrintCellTextureCache), because the preview is where the tools are used: window/level, zoom, pan, rotate, flip and invert were each a CPU re-render per mouse delta. The arrangement is the display shader's transform, so those drags now re-draw a quad and render nothing; only a window change makes a new texture, and that is one GPU dispatch. A re-windowed cell holds its previous texture while the new one renders, so a drag never falls back to the CPU. Falls back per cell — overlay planes, YBR, an unresolvable window, no Metal — never per film. Only the film on screen holds textures.

• The shader is given the film's geometry, not the viewer's: new DisplayPresentation.sourceRegion fits and centres the region that ViewerPresentation.visibleRegion — the same call the print path makes — says will be cropped, and flips after the rotation as PrintPresentationTransform does. The viewer's own transform agrees with the printer only while a cell is merely zoomed, so reusing it would have misreported every rotated or edge-panned cell.

• Film composition and export are unchanged and still CPU, as the GPU plan fences them: preview and film agree because neither invents anything the other does not.

• On macOS the print preview opens in its own window (StudioWindowID.printPreview) instead of a modal sheet over the viewer, so the film can be compared with the images it was made from, moved to a second display, or zoomed to fill one. ⌘P and the library's "Print…" both raise it; PrintScreenPresenter keeps the sheet on platforms without windows.

• PrintSettingsView takes a presentation (.sheet / .window): a sheet is still given a fixed size, a window only a minimum, so the user's own size sticks.

• The Print screen gained a "Print Preview" button that raises the window — the preview no longer has to be re-opened from the viewer once it has been closed.

• Opening a study closes the print preview window along with clearing the marks: the film on it was composed from the study being left. prepareForNewStudy() bumps printScreenDismissRequests, which the shell turns into a dismissWindow — watched there rather than in the viewer, since the study may be opened while the library is still on screen.

• The print log console starts closed and opens by itself when a job starts, giving the film its width while the film is what is being judged. It is put away again on "Print Again", and the header toggle still overrides both.

Added — Metal GPU frame rendering, DICOMRenderKit (2026-08-04)

Full GPU rendering pipeline for the viewer, landing GPU plan milestones M0–M7 (see GPU_RENDERING_PLAN.md):

• New DICOMRenderKit target: a Metal compute pipeline that windows/LUTs a decoded frame directly into a CGImage-readable texture with zero-copy UMA input and no upload/render/readback round trip. Monochrome and RGB/palette kernels; YBR stays on the CPU. Output is byte-for-byte identical to the CPU path (MetalCPUEquivalenceTests), so the two are freely interchangeable.
• Shared WindowLUT (integer, not float shader math) backs both the CPU and GPU paths — 10–19× faster monochrome rendering on its own (M0+M1), before any GPU work.
• Focused-viewport direct-to-display path keeps the frame on the GPU across tool actions (window/level, invert, zoom, pan): 0.008 ms per action, down from a full re-render (M5). The per-drag decode is cached too, up to 675× faster per step.
• minimumGPUPixelCount dropped from 1 megapixel to 0 — no frame is declined for being small; the CPU is now purely the no-GPU fallback, not a size-selected alternative. Safe only because of the CPU/GPU output equivalence guarantee above.
• Shipped the compute shader as .metal.txt so the app builds without requiring the separate Metal Toolchain install in Xcode.

Added — corner annotations and GPU textures for unfocused tiles (2026-08-04, in progress)

Work in progress, not yet committed.

• On-screen viewer tiles and the focused viewport now carry the traditional four-corner reading-room annotation layout — size/window/cursor readout (top left), patient/study identification (top right), zoom/position/ compression/geometry (bottom left), acquisition date/time (bottom right) — composed by ViewerAnnotationCorners/ViewerAnnotationText and drawn by the new ViewerAnnotationOverlayView. Detail scales down (.full → .reduced → .minimal) as a tile shrinks, keeping identification and position and dropping the rest rather than shrinking everything to illegibility.
• ViewerHoverGeometry maps a cursor point back through the viewport's fit, zoom, pan, rotation and flip to the underlying image pixel — separately for the CPU/print transform order (pan before rotation) and the GPU display shader's order (pan after rotation), since they only agree when a picture is unrotated or unpanned — so the top-left corner can show the pixel value and patient-space position under the cursor. A wrong readout is worse than none, so every step returns nil rather than guess when the point is off the picture.
• PatientIdentificationOverlayView is now film/print-only — the on-screen viewer's old single-band overlay is replaced by the corner layout above, since on screen the reader can move the picture out from under the text but paper cannot.
• New ViewerTileTextureCache extends the GPU display path (M5) to unfocused grid tiles: a texture keyed only on file + frame + window, so a synchronized zoom or window drag across a grid re-draws a quad per tile instead of re-rendering one. This is the GPU-tile work M6 explicitly deferred (" ViewerTileImageCache ... stay on the readback path") — tiles now have both a CPU image and, where Metal is available and the frame supports it, a GPU texture, falling back to the CPU image per tile (overlay planes, an unresolvable window, no Metal device) rather than per grid.
• Also removed the unused Inspector-panel toggle from MainView/MainViewModel (dead code, unrelated cleanup).

Removed — CLI-parity test harness (2026-08-03)

Deleted the whole Tier-2 CLI-parity subsystem: the "CLI Parity" Studio screen and its engine/comparators, the orphaned CLI Automation Testing screen, bundled synthetic fixtures and goldens, the cli-parity-gen/cli-parity-docs/studio-cli-introspect dev targets, all CLIParity*Tests suites plus the StudioParityTests target, the "Tier-2 Output Parity Gate" CI job, and every APP_CLI_*PARITY*/docs/cli-parity/ doc — 103 files, ~14,400 lines of code, ~5,100 lines of docs, 1.1 MB of fixtures.

• Kept deliberately: CLIToolTerminalCompare + CLIToolBuilder (reachable only from code now that the Workshop's comparator panel is gone) and the syn-ct.dcm fixture the print tests depend on, now Tests/DICOMStudioTests/Fixtures/syn-ct.dcm via a new StudioTestFixtures helper.
• APP_CLI_SHARED_API.md updated to describe verification as it stands today: the oracle-based round-trip suite (DICOMRoundTripTests) plus, per tool, an optional shared *Console type.

Added — shared consoles for dicom-split/merge/script (2026-08-03)

dicom-split, dicom-merge and dicom-script were the last three tools whose Workshop executors shared only the engine while duplicating console text and input parsing. Text and parsing now live once in DICOMKit — SplitConsole, MergeConsole, ScriptConsole — pinned by Tests/DICOMRoundTripTest/SharedConsoleParityTests.swift. Fixed real drift found in the process: the app's banners had dropped the CLI version string, dicom-merge relabelled its input-count line and echoed --format values as enum case names instead of the CLI's raw text, and dicom-split's frame parser silently accepted whitespace-only components the CLI rejected.

Fixed — Print SCP honors SCU-supplied SOP Instance UIDs (2026-08-03)

A dcm4che-based Print SCU failed at N-ACTION with 0x0112 "Unknown Film Session" and never produced a film. PS3.7 10.1.5 lets the SCU supply the Affected SOP Instance UID on N-CREATE, and such SCUs then address every follow-up N-SET / N-ACTION by their UID regardless of what the response carries — while our SCP minted and stored its own.

• PrintSCP.assignedUID(requested:) stores the object under the SCU's UID whenever one is supplied, for Film Session, Film Box, Presentation LUT and Basic Annotation Box N-CREATE; it falls back to the generator otherwise. A supplied UID must be non-empty after trimming NUL/space padding, ≤ 64 characters, and digits-and-dots only, so a malformed value can never become a stored key.
• Film Box N-CREATE now rejects a UID already in use with 0x0111 (Duplicate SOP Instance) — unreachable while the SCP minted its own.
• Image Box UIDs remain SCP-allocated; they are created implicitly with the film box.
• PrintSCPStatusMatrixTests.testSCUSuppliedSOPInstanceUIDsAreHonored replays the field sequence end to end and asserts the composed film carries the SCU's UIDs.
• PRINT_CONFORMANCE.md §2.1 / §4 and DICOM_PRINT_SCP_PLAN.md updated.

Added — DICOM Print SCP: emulator screen and dicom-printscp CLI (2026-07-31)

Milestones E and F of DICOM_PRINT_SCP_PLAN.md — the last two rows in that plan's matrix. The app surface came first, so the sharing ran in the direction opposite every other print entry: settings, assembly and wording were moved out of DICOMStudio into DICOMPrintKit, and both the Studio screen and the CLI now consume rather than own them.

• PrintSCPSettings / PrintSCPService / PrintSCPConsole / PrintSCPSimulator (DICOMPrintKit/Printing/): every configuration knob and its PrintSCPConfiguration / FilmComposerConfiguration mapping, the sink stack and server assembly, the wording of every event/film/startup line, and a no-network film simulator, in one place — Studio's view model now owns only what a window owns (retained films, selection, button state).
• dicom-printscp (Sources/dicom-printscp/): serve (default) / simulate / status / queues. simulate takes DICOM files rather than a film.json descriptor as originally sketched — the composer needs real pixels, and routing through PrintImagePreparer means a simulated sheet is built by the same code a live SCU's job is.
• Studio's "Printer Emulator" (PrintSCPView, PrintSCPViewModel, PrintSCPSettingsStorageService): a persistent Window (not a WindowGroup — one emulator, one listener) reachable from the sidebar while a print is in flight from the main window.
• Two defects a live SCU→SCP run caught, both fixed in the shared core: --max-films counted off the screen stream (delivered the instant a sheet is composed) instead of .filmPrinted, so the listener could stop mid-N-ACTION and abort the SCU's association before its PNG was written; and the association log line rendered the port as :0 because AssociationInfo.remotePort is always 0 for this SCP and the whole endpoint is in remoteHost — fixed once in the shared console.

Plan: DICOM_PRINT_SCP_PLAN.md ("Milestone F as built" section has the full shared-type table). Not in the CLI-parity harness (print tools are covered by dedicated tests instead, per APP_CLI_SHARED_API.md).

Added — DICOM Studio: overlay planes, patient identification, and print window fixes (2026-07-31)

• Overlay plane rendering (DICOMKit/OverlayPlaneRenderer.swift): PS3.3 C.9.2 group-60xx overlay bitmaps, read and drawn both as a CGImage composite (viewer) and burned into raw samples (film). Written for Siemens' "Patient Protocol" Secondary Capture, whose Pixel Data is entirely zero and whose whole content lives in a 1-bit overlay — previously rendered as a black square, now shown/printed correctly. Wired into both ImageViewerViewModel (render path) and PrintImagePreparer (so a film matches the screen it was approved on, except under --raw).
• Print window resolution now falls back to the file's own VOI (PrintImagePreparer.resolvedWindow, PrintWindowSpace): a mark made without opening the file (whole-series or library print) previously auto-stretched a CT's full pixel range and left soft tissue in a handful of indistinguishable greys; it now falls back to the data set's VOI window like the viewer, export and tiles already do. PrintWindowSpace (.outputUnits / .storedValues) travels with an explicit window so a value taken off the viewer (stored units) is converted through Rescale Slope/Intercept before printing — sending it unconverted put the window entirely outside the pixels and printed a flat black cell.
• Arrow geometry fully consolidated (PrintArrowGeometry.swift): the fractional-of-image-height math for shaft width and head size, previously duplicated between ImageAnnotationBurner (film) and FilmCellAnnotationLayer (preview), now lives once and both call it; the preview's arrow rendering also switched from a shadow-based halo to a stroked-outline halo so the head no longer blurs into the shaft and reads as a plain line. Selection handles on a selected arrow shrank to a small ring with a separately-sized (2x) invisible grab area, so a handle no longer hides the anatomy the arrow points at.
• Viewer patient-identification header (ImageViewerViewModel+PatientOverlay): patientIdentityLine ("name, ID · CT / PT"), modalitiesForOverlay (every modality in the open study, series order, no repeats — a study is routinely PET/CT or a CT with an SR), and studyDescriptionSanitizedForOverlay (strips the caret/pipe/backslash separators a description arrives with, keeping only what reads as words).
• Viewer toolbar reworked around click-armed tools: windowing and zoom are now toggled tool buttons (highlighted while armed) rather than ⌥-drag/⌘-drag modifiers, freeing ⌘-drag; resetView() now also resets window/level and inversion, not just pan/zoom/rotation; the inline "Open DICOM File" button, file importer and ⌘O shortcut were removed from the viewer.
• Tests: PrintSCPSharedCoreTests (23), PrintSCPScreenTests (39), PrintWindowSpaceTests (9), OverlayPlaneRendererTests (17), PrintMarkWindowTests (5), ViewerIdentificationHeaderTests (5); PrintOverlayAnnotationTests, PrintCellEditingTests, NavigationServiceTests, ViewerTileLayoutTests and PolishReleaseViewModelTests updated for the arrow-geometry, cell-editing-window and toolbar/reset-view changes.

Added — DICOM Studio: the film as a working surface (2026-07-30)

Plan: DICOM_PRINT_STUDIO_PLAN.md §10.

• Patient identification burned into the film (DICOMPrintKit/ImageAnnotationBurner.swift, PatientOverlayText, PatientOverlayTextCache, PatientIdentificationOverlayView): "name, ID, study date" over the study description, from one definition shared by viewer tiles and film cells. A DICOM printer draws identification from annotation boxes to its own layout and many ignore them, so the lines are burned into the prepared 8-bit frame instead — one bitmap pass per cell, applied last so a later crop or rotate cannot carry the caption with it, and skipped quietly on an unexpected pixel format rather than failing the print. The viewer draws it as a reserved band below the picture; the film preview overlays it, since there the cell is the film.
• Drawn annotations on a film cell (PrintOverlayAnnotation, PrintViewModel+Annotations, FilmCellAnnotationLayer): text and arrows, in coordinates normalized to the image and held per mark ID — so re-arranging the film, changing layout or printing to another sheet cannot move an arrow off the vessel it pointed at, and the 512-pixel preview and the 3000-pixel frame agree. The preview uses the burner's own arrow geometry.
• Per-cell editing in the print preview (PrintViewModel+CellEditing): window/level, zoom, pan, text and arrow tools writing back into the mark that PrintService.prepare reads, so the preview cannot drift from the film. FrameSourceCache keeps decoded pixels so a window/level drag re-maps rather than re-decoding (tens of ms per event on a JPEG 2000 CT); cell renders moved 256 → 512 (at 256 half the grey levels being judged are lost to downscaling) and the previous rendering stands in during a gesture instead of a spinner.
• Non-image instances open instead of failing (ViewerContentKind, ViewerNonImageContent, ViewerNonImageContentView): SR read as a narrative, encapsulated PDF as pages, presentation states / KOS / raw data as named summaries. Previously a valid SR was reported as an "unsupported transfer syntax" — a decode failure for pixels it never claimed to have.
• Library rows describe the study again (StudyModel.merging, StudyRowSummary, LibraryModel, DICOMFileService): study fields are unioned across the files of a study rather than overwritten by whichever was read last; rows fall back to the series' modality and description, then patient ID / accession, before saying "Unknown"; series and instances sort by Series/Instance Number with unnumbered last and a total-order tiebreak (a series used to open on a different image on each read); empty studies are pruned; a DICOMDIR and any object with no SOP/Series Instance UID are refused at import instead of manufacturing an unopenable "Unknown Patient, 0 series, 0 images" row.
• Reading with one hand on the keyboard: shared ScrollWheelHandler (viewer pages images, film preview zooms cells, both live while the print sheet is over the viewer) with step accumulation for fine trackpad deltas; first/last jumps and opt-in wrap at series ends; ViewerPrintTrayView — the selection in film order, each row rendered with that mark's own window and arrangement, without opening the print sheet; ⌘/ keyboard-shortcut legend on both screens.
• Tests: PrintOverlayAnnotationTests, ImageAnnotationBurnerTests, PrintCellAnnotationTests, PrintCellEditingTests, PatientOverlayTextTests, PatientIdentificationOverlaySizingTests, ViewerContentKindTests, ViewerProtocolLineTests, LibraryStudyMergeTests, LibraryImportEndToEndTests, StudyRowSummaryTests, ScrollStepAccumulatorTests, ViewerTileLayoutTests — 4,335 tests in 394 suites green.

Added — DICOM Print SCP: printer emulator (2026-07-28/29)

DICOMKit now implements Print Management (PS3.4 Annex H) in both roles. Any Print SCU — modality, workstation or third-party tool — can associate, send a film session, film boxes and image boxes, and issue N-ACTION print; the film is composed and handed to an output sink. Conformance statement: PRINT_CONFORMANCE.md. Plan: DICOM_PRINT_SCP_PLAN.md.

• Protocol machine (DICOMNetwork): DICOMPrintServer actor + PrintSCP.swift, PrintSCPTypes.swift, PrintSCPEncoder.swift — association accept/reject, called/calling AE checks, N-CREATE / N-SET / N-GET / N-ACTION / N-DELETE / N-EVENT-REPORT across Film Session, Film Box, Grayscale/Color Image Box, Printer, Print Job, Presentation LUT and Basic Annotation Box, with the PS3.4 H.4 lifecycle and cascade deletes. Configurable via PrintSCPConfiguration (film sizes, medium types, colour, max boxes per film, annotation boxes, idle timeout, AE lists, printer identity).
• SCP-direction dataset parsing (PrintDatasetReader.swift, PrintSCPParser.swift): a real element walk for both Explicit and Implicit VR LE, including sequences with defined and undefined length and image-box PixelData, mapped back onto the existing FilmSession / FilmBox / ImageBoxContent / PrintImageData model.
• PrintImageDisplayFormat: parses STANDARD / ROW / COL / SLIDE / SUPERSLIDE / CUSTOM, driving image-box allocation, composer cell layout and the SCU's box maths.
• Film composition (DICOMPrintKit/Printing/): FilmGeometry (sheet sizes at a configurable DPI, cell layout, fitting), FilmComposer (image boxes → one page bitmap: magnification, polarity, LUT shape, border/empty density, trim marks) and ComposedFilm. Four inversions compose correctly — MONOCHROME1, Polarity REVERSE, INVERSE / LIN OD Presentation LUT shape, and film emulation. DensityMapping is .paperDirect by default, with .filmEmulation opt-in rather than a silent guess.
• Output sinks: PrintOutputSink protocol with ScreenSink (full-resolution stream + bounded downsampled scrollback; never blocks the SCP on a viewer that is not draining), PDFSink, ImageSink (PNG/TIFF), PaperPrinterSink (CUPS lp, opt-in) and CompositePrintSink; FilmComposingPrintHandler wires the SCP to them.
• Tests: PrintSCPTests.swift (74 — parser round-trip in both VRs, encoder, loopback SCU → SCP, status matrix, robustness) and DICOMPrintKitTests (63 — geometry, composer, sinks, screen scrollback, DCMTK interop).

Fixed — DICOM Print interoperability, verified against DCMTK 3.7.0 (2026-07-29)

Automated in Tests/DICOMPrintKitTests/DCMTKInteropTests.swift (skips when DCMTK is absent): dcmpsprt/dcmprscu → our SCP, and our SCU → dcmprscp (IHE Full profile).

• SCU proposes the meta and the individual SOP Classes and routes each N-service to an accepted context (PrintPresentationContexts). A printer that rejects the meta class was previously unusable.
• Image Display Format order was a conformance bug: PS3.3 C.13.3 defines STANDARD\C,R as columns-first; PrintLayout.imageDisplayFormat emitted rows-first, so every non-square layout printed transposed on a conformant printer. PrintLayout is now the single source of the string.
• Configuration Information (2010,0150) threaded end to end — Studio and the CLI collected it but PrintOptions had no field and it was never sent; several vendors require it.
• 0x0106 reclassified as a failure (DIMSEStatus) — it was treated as a warning, so a job sailed past a printer's rejection and failed later with a misleading code.
• Trim (2010,0140) omitted when NO and Requested Decimate/Crop Behavior (2020,0040) omitted when DECIMATE — printers that do not implement them reject a box for merely carrying them.
• SCP idle-association timeout (default 300 s) so a vanished peer no longer holds a slot forever, and A-ASSOCIATE-RJ at capacity instead of dropping the socket.
• Error Comment (0000,0902) transliterated to ASCII — CommandSet encodes as ASCII and silently dropped any value containing a non-ASCII character, discarding the only diagnostic an SCU developer gets.

Added — DICOMPrintKit: shared print core for the CLI and Studio

• New target/product DICOMPrintKit (DICOMCore + DICOMDictionary + DICOMKit + DICOMNetwork), consumed by dicom-print and DICOMStudio. The shared core cannot live in DICOMKit — it needs both ImagePreprocessor and PrintImageData — and putting networking into DICOMKit would force it on all ~30 CLIs.
• PrintJobRequest (every job knob as one value type, plus validate(), filmCount(for:) and PrintPlan), PrintOptionCatalog (one table of selectable values per option, feeding both UI pickers and the CLI's arg enums), PrintImagePreparer, PrintWorkflow (preflight, retries, events, progress, job/printer status) and PrintConsoleFormatter (text + JSON).
• dicom-print refactored onto all of the above — arg declarations, help text and output unchanged; PrintCLIEndToEndTests, PrintServiceTests and PrintSCPIntegrationTests stay green.
• Multi-film gap fixed: PrintResult now carries filmBoxUIDs and printJobUIDs (the singular properties remain as last-element accessors), so per-film job-status polling and job history work for multi-film jobs.

Added — DICOM Studio: print workflow (2026-07-28)

Library → viewer → mark images → print. Plan and full file map: DICOM_PRINT_STUDIO_PLAN.md.

• Marking in the viewer: ordered, frame-level PrintSelectionModel; M marks the current image, ⌘P opens the print sheet, a badge shows the count, a capsule shows the film position, and the context menu carries mark-all-frames / whole-series / clear.
• Print settings sheet: printer picker with Test (C-ECHO) and Status (N-GET), layout (auto | grid | preset), film size, orientation, copies, a live film preview showing spillover, a marks tray, and an Advanced disclosure covering the rest of the CLI surface. (Reworked 2026-07-29 — see below.)
• Printer management: PrinterProfile + PrinterProfileStorageService (printer-profiles.json), modelled on the existing PACS server profiles. The CLI's ~/.config/dicomkit/printers.json registry stays independent by design — a sandboxed app cannot reach it.
• Execution: PrintViewModel + Studio PrintService over DICOMPrintKit, determinate per-image-box progress, live N-EVENT-REPORT console, cancel with film-session cleanup, per-film Check Status, and job history persisted to print-job-history.json.
• New Print navigation destination (printers + job history) and a Print… action on a study/series in the library, which adds those files to the selection without discarding marks already made in the viewer.

Fixed — DICOM Studio: windowing, series order and the print sheet layout (2026-07-29)

• Multi-valued VOI no longer discarded (DICOMImageExporter.determineWindowSettings): a CT that carries a lung and a soft-tissue window in one element (-600\50 / 1200\350) fell through windowSettings() — which parses a single DS — and was auto-stretched over the full pixel range. The multi-valued form is read first and its first pair, the default presentation, wins.
• One window policy everywhere: ImageViewerViewModel now adopts a file's default window through determineWindowSettings (the same call the exporter, tile cache and film use), and FrameRenderer's fallback ladder goes through it too instead of renderFrameWithStoredWindow, which hands a raw HU centre to a renderer reading stored values and washes a CT with a large Rescale Intercept out to white.
• A window no longer follows the user across a hang: ViewerCellState.windowCenter/Width are optional, nil meaning "this image's own VOI". A freshly hung tile starts at its own VOI, and the viewer's window is inherited only by tiles in the same series — filling a grid from a flat file list used to stamp one kernel's window across an MPR and every other reconstruction in the study.
• Series pane in series-number order: ascending, unnumbered series last (the library's ordering treated a missing number as 0, filing them ahead of series 1), ties broken by title then UID so the pane cannot reshuffle between two reads. Cards show the series number badge, and VoiceOver announces "Series 4, …".
• Print sheet reworked: options moved from a 320 pt left column into a band across the top so the film preview owns the whole centre; the sheet opens at the size of the window it was raised from; Advanced expands sideways under a height cap instead of pushing the film off the bottom; the marks tray became a "Show List…" popover, since film order now follows the viewer's tile order (syncPrintOrderToViewer) rather than the order the boxes were ticked.

Added — DICOM Studio: the film matches the screen (2026-07-28)

• ViewerPresentation + PrintPresentationTransform (DICOMPrintKit): a mark now records the viewer's arrangement as geometry over the source image — zoom, pan, viewport, quarter turns, flips, invert — and the print path crops, permutes and negates the full-resolution frame accordingly. Nothing resamples, so a zoomed print carries the modality's real detail rather than an upscaled copy of what the monitor showed.
• Film preview shows the actual frames: FrameRenderer + FrameImageStore + PrintThumbnailCache render each marked frame as it will print. Cache keys include the whole arrangement, so two marks of the same frame at different zooms are two pictures.
• ImageInversion: the viewer inverts the rendered frame rather than negating the VOI window, which is not equivalent once Rescale Slope/Intercept or a signed representation are involved; the print path inverts P-values directly.

Added — DICOM Studio: viewer tile grid and series pane (2026-07-28)

• Tile grid 1×1 … 4×4 whose cells map to film cells in the same order. The focused tile is the live view model, so gestures, window/level and cine behave exactly as at 1×1; every tile keeps its own file, frame, window, zoom/pan, rotation, flips and inversion. Unfocused tiles are cached renders keyed on full tile state, so panning one tile does not re-decode the others.
• Series pane: every series of the open study as a card (thumbnail, description, objects and frames, current/visited state). Drag a card onto a tile, or select a tile and double-click, to hang a different series there. Built from indexed library metadata, with orientation read from one file per series afterwards and folded in.
• Arrow-key navigation by image, not by file: frames first, rolling onto the neighbouring file and stopping at the end of the series; wrapping stays cine behaviour.
• Tests: ViewerPresentationTests, PrintPresentationTransformTests, PrintViewerPresentationEndToEndTests, PrintThumbnailCacheTests, PrintSelectionModelTests, ViewerTileLayoutTests, ViewerSeriesPaneTests — DICOMStudioTests 4,185 tests green.

Fixed — DICOMNetwork test-suite hang and unmasked failures (2026-07-27)

• CommitmentNotificationListener.waitForResult hang: the timeout race used a task group whose waiter child suspended in a non-cancellation-aware continuation, so after the timeout threw the group could never drain — testCommitmentNotificationListenerWaitForResultTimeout (and any caller hitting the timeout path) hung forever, stalling full DICOMNetworkTests runs after ~880 tests. The wait is now a single continuation registered synchronously on the actor and resumed by exactly one of: result arrival, timeout, or stop(). stop() also no longer waits on a listener that is already cancelled.
• ClientIdentity keychain lookup: kSecClassIdentity queries do not reliably filter on kSecAttrLabel, so a lookup for a non-existent label could return an arbitrary keychain identity (wrong client certificate). The lookup now fetches attributes for all candidates and matches the label explicitly, throwing keychainIdentityNotFound when nothing matches.
• StoreAndForwardQueueTests.test_queue_enqueueDrainingThrows race: an empty queue could finish draining (→ .stopped) before the test's enqueue ran; the test now holds the queue in .draining via notifyConnectivityLost() first.
• Full DICOMNetworkTests (1086 XCTest + 192 swift-testing tests) now completes green in ~15 s and can gate CI.

DICOM Print — post-plan pending work (items 1–7)

• Palette color printing: PALETTE COLOR sources are mapped through the data set's Red/Green/Blue palette LUTs to RGB (or luminance grayscale); a missing LUT module produces a clear error instead of notYetImplemented.
• Uncompressed subsampled YBR: packed YBR_FULL_422 (full range) and YBR_PARTIAL_422 (BT.601 studio range) frames are chroma-upsampled and converted to RGB; 4:2:0/ICT/RCT remain rejected (never occur uncompressed).
• Deep grayscale output: --bit-depth 8|12|16 — depths above 8 emit little-endian 16-bit-allocated P-Values with matching Bits Stored/High Bit.
• Explicit VOI window: --window-center/--window-width override the data set's window.
• Full Implicit VR LE support: print presentation contexts propose Explicit VR LE with Implicit VR LE fallback, and both serialization and all response parsers honor the negotiated syntax — implicit-only printers now work end-to-end (verified against the mock SCP in both syntaxes).
• --magnification none exposed (library case existed).
• Spawn-based CLI end-to-end tests: PrintCLIEndToEndTests runs the built dicom-print binary — version/validation exit codes, dry-run behavior, the JSON stdout contract, and a full print + failure path against the in-process mock Print SCP.

DICOM Network — release-window P-DATA tolerance (print plan P2-7)

• Association.release() no longer aborts on a late P-DATA PDU: a message pushed by the peer between A-RELEASE-RQ and A-RELEASE-RP (e.g. a Print SCP's N-EVENT-REPORT) previously hit the unexpected-PDU path — abort + error on an otherwise successful operation. Per PS3.8 §7.2 the release requestor now discards P-DATA received in the release window and keeps waiting for A-RELEASE-RP. Integration-tested with the mock Print SCP.

DICOM Print — Milestone D: CLI re-enabled + mock SCP integration tests

• dicom-print re-enabled in Package.swift (product + executable target, owner approved) — previously excluded under Phase-1 scope. Builds with zero warnings.
• Mock Print SCP test harness: new in-process, NWListener-based MockPrintSCP (DICOMNetworkTests) implementing A-ASSOCIATE accept/reject, N-GET printer status, N-CREATE film session / film box (Referenced Image Box Sequence sized from the requested Image Display Format), N-SET, N-ACTION, N-DELETE, and A-RELEASE — with scriptable failure injection (status + Error Comment/ID), silence-after-accept, presentation-context rejection, omitted job UID, and pushed N-EVENT-REPORTs.
• 10 end-to-end workflow tests (PrintSCPIntegrationTests): happy path with exact DIMSE sequence + single-association assertion (PS3.4 H.4), multi-film on one association, failure injection carrying "OUT OF FILM (Error ID 42)" to the thrown error, defensive cleanup, silent-SCP timeout, zero-context rejection, interleaved event delivery + acknowledgement, omitted-job-UID handling, and printer-status round-trip.
• Defensive cleanup on failure (P2-3): when a later workflow step fails, the SCU now attempts a best-effort in-association Film Session N-DELETE before aborting (the inner guards no longer abort pre-throw, so the association is still alive for cleanup).
• Machine-readable output contract (P3-2): send gained --format json (result object on stdout); the stdout/stderr/exit-code contract is documented in the CLI README.

DICOM Print — Milestone E hardening + CLI ergonomics (partial)

• Scoped image-box UID parsing (P2-2, bug fix): parseImageBoxUIDs scanned the whole response for (0008,1155) and could mis-attribute annotation-box or presentation-LUT references as image boxes; it is now bounded to the Referenced Image Box Sequence (2010,0510).
• Empty Print Job UID no longer recorded (P2-4): the workflow silently appended an empty job UID when the N-ACTION response omitted it; consistent with the discrete printFilmBox, empty UIDs are now dropped.
• Port bounds guard (P2-5, crash fix): pacs://host:99999 trapped on UInt16 conversion; now a clean validation error.
• New send options (P3-1): --magnification replicate|bilinear|cubic, --film-destination magazine|processor|bin-1|bin-2, and the full film-size set (8.5x11, 24x24cm, 24x30cm added).
• Pre-flight checks: --check-status (P2-1) queries printer status before printing — aborts on FAILURE, warns on WARNING; --verify (P3-3) performs a C-ECHO connectivity check first.

DICOM Print — Milestone C (interoperability) from the enhancement plan

• Image-box pixel attributes always sent (P1-1, bug fix): the N-SET Preformatted Image Sequence previously omitted Rows/Columns/BitsAllocated/ PhotometricInterpretation when no descriptor was supplied — rejected by strict SCPs. The print workflow now requires one PrintImageData per image (validated up front, before any network activity) and emits the attributes unconditionally; the discrete setImageBox throws a clear error without a descriptor.
• printWithTemplate / printImagesWithProgress on a single association (P1-2, conformance fix): both previously opened a separate association per DIMSE step (PS3.4 H.4 violation — the Film Session UID does not survive across associations). Both are reimplemented on the single-association workflow; the progress stream keeps its phase/percent updates via a new internal progress hook, and both gained imageDescriptors:/eventHandler: parameters.
• Transfer syntax (P1-3, documented decision): print presentation contexts now propose Explicit VR LE only. Previously Implicit VR LE was also proposed but data was always serialized Explicit — an implicit-only SCP would accept a syntax we then mis-encoded. Now such an SCP cleanly rejects negotiation; full Implicit-VR support remains future work if a real printer needs it.
• DIMSE-response timeout (P1-4, bug fix): an SCP that accepted the association and then went silent hung the tool forever. All print DIMSE response reads now race against PrintConfiguration.timeout; on expiry the association is aborted and operationTimeout is thrown.

DICOM Print — Milestone B (image fidelity) from the enhancement plan

• Preprocessing pipeline wired into printing (P0-1, bug fix): dicom-print send now runs every frame through ImagePreprocessor by default — Rescale Slope/Intercept → VOI window (from the data set or auto-calculated) → MONOCHROME1 inversion → 8-bit MONOCHROME2 output (8-bit RGB for color mode). Previously raw stored pixel values were sent, so windowed CT/MR and MONOCHROME1 images printed with clinically incorrect grayscale/polarity. A --raw flag bypasses the pipeline. The two XCTSkip-quarantined MONOCHROME1 preprocessor tests were rewritten to the decided behavior and re-enabled.
• Encapsulated pixel data decoded before N-SET (P0-2, bug fix): compressed sources (JPEG, JPEG 2000, JPEG-LS, RLE) were shipped as raw encapsulated fragments — malformed image boxes. send now decodes to native frames via DICOMFile.tryPixelData(), which also applies the JPEG-Baseline YBR→RGB descriptor correction.
• Multi-frame handling (P0-6, bug fix): previously the entire multi-frame Pixel Data value was sent as one image. New --frame N (1-based, default 1) and --all-frames (one image box per frame) options with bounds validation.
• YBR color conversion (P1-5): uncompressed YBR_FULL sources are converted to RGB for color printing (PS3.3 C.7.6.3.1.2); RGB→grayscale conversion is applied for grayscale mode. Uncompressed subsampled YBR (YBR_FULL_422 etc.) is explicitly rejected with a clear error rather than mis-converted (packed 4:2:2 layouts need bytesPerFrame modeling in DICOMCore first).
• Signed pixel safety (P2-6): with preprocessing on by default, Pixel Representation = 1 sources are sign-extended and emitted as unsigned 8-bit P-Values — signed values are no longer sent to unsigned Image Boxes.
• New ImagePreprocessor.prepareForPrint(pixelData:dataSet:frameIndex:colorMode:) API prepares a single frame of already-decoded pixel data (the existing data-set variant now delegates to it).

DICOM Print — Milestone A (safety) from the enhancement plan

• Real printer-status parsing (P0-3, bug fix): parsePrinterStatus was a stub that always returned NORMAL regardless of the N-GET response, so PrinterStatus.isNormal was always true. It now decodes Printer Status (2110,0010), Printer Status Info (2110,0020), Printer Name (2110,0030), and — when returned — Manufacturer (0008,0070) / Manufacturer Model Name (0008,1090). A response without a status attribute now reports "UNKNOWN" instead of a false NORMAL. dicom-print status surfaces the new fields in text and JSON output.
• Non-zero exit code on print failure (P0-4, bug fix): dicom-print send now exits with a failure code when the print result is unsuccessful; previously it printed "✗ Print failed" but exited 0, so automation could not detect failures.
• DIMSE error detail surfaced (P0-5): CommandSet gained errorComment (0000,0902), errorID (0000,0903), and offendingElements (0000,0901) accessors. DICOMNetworkError.printOperationFailed now carries an optional detail string populated from the SCP's Error Comment / Error ID on every print failure path, so users see e.g. "OUT OF FILM" instead of only a numeric status. A one-argument printOperationFailed(_:) factory preserves source compatibility.

DICOM Print tool (dicom-print) — features + fixes

• --layout now honored (bug fix): the flag was parsed and echoed but never applied — send always used the automatic layout. printImages gained an optional explicit layout: override (nil = existing auto-layout) and the CLI now threads --layout through. A single image with an explicit --layout is routed accordingly.
• --color added (gap fix): the CLI always printed grayscale because PrintConfiguration.colorMode was never set. Added --color grayscale|color on send, which negotiates the Color Print Management Meta SOP Class and sends color image boxes.
• Conformant image descriptors: send now extracts per-image attributes (rows, columns, bits allocated/stored, high bit, samples per pixel, pixel representation, photometric interpretation) from each dataset and sends them in the N-SET Preformatted Image Sequence (PS3.3 C.13.5.1). Previously required image attributes were omitted.
• N-EVENT-REPORT reception: the Print SCU now receives, decodes, and acknowledges asynchronous printer/print-job notifications pushed by the SCP (Printer SOP Class status; Print Job SOP Class progress). New PrintEvent, PrinterEventType, PrintJobEventType, and a PrintEventHandler callback wired through printImage/printImages. dicom-print send prints faults always and routine progress in --verbose. This also fixes a latent correctness bug where an interleaved event could be mis-parsed as the awaited DIMSE response.
• Build: the dicom-print target was excluded from Package.swift and had drifted out of compilability (DICOMParser/data(for:) no longer existed, @Sendable capture errors). Repaired to use DICOMFile.read(from:force:).
• Layout presets + retry: dicom-print send gained --template (single/comparison/grid/multi-phase — sets layout + film size + orientation, routed through the conformant single-association path) and --retries N (retry on connection/setup failure with exponential backoff; a submitted job is never retried, so no duplicate prints).
• Presentation LUT: added PresentationLUTShape and a presentationLUTShape print option. When set, the workflow N-CREATEs a Presentation LUT SOP Instance (part of the Grayscale/Color Print Management Meta, so no extra presentation context) and references it from each film box (Referenced Presentation LUT Sequence, 2050,0500). CLI: --presentation-lut identity|inverse|lin-od.
• Annotation boxes: added PrintAnnotation and annotations / annotationDisplayFormatID print options. The workflow sets Annotation Display Format ID on the film box and N-SETs each Basic Annotation Box (position + text) using a new sequence-scoped UID parser so annotation-box UIDs are not confused with image-box UIDs. CLI: repeatable --annotate <text> + --annotation-format <id>. Note: the Annotation Display Format ID is printer-specific.
• Overlay box scaffolding: added the Basic Print Image Overlay Box SOP Class UID and Referenced Image Overlay Box Sequence tag; full overlay-plane extraction remains a follow-up.

Tests

• Re-enabled the DICOMNetworkTests target (was excluded from Package.swift), so print logic — including the new Presentation LUT / annotation / N-EVENT-REPORT code — is covered again (177 tests). The rotted, live-PACS PACSIntegrationTests is quarantined via exclude: until ported to the current API; two outdated MONOCHROME1 ImagePreprocessor expectations are XCTSkip-quarantined pending a product decision on 8-bit vs 16-bit print output.

Fixed — Bug review pass (library crashes/correctness + CLI hardening)

• STOW-RS server (critical): the multipart parser decoded the entire body as UTF-8 before splitting, so binary application/dicom uploads (almost never valid UTF-8) parsed to zero parts and were silently dropped while the server still reported success; rare bodies that did decode were corrupted by whitespace-trimming raw bytes. Now uses the same byte-scanning MultipartMIME parser as the client. See BUG_REVIEW.md (C1).
• SIMD window/level: applyWindowLevel passed a positive offset instead of -minValue into the vDSP scalar-add, blowing out every image rendered through the fast path. PerformanceTests/SIMDImageProcessorTests.swift had also silently dropped out of the build (excluded in Package.swift), so the regression test for this never ran; both are now fixed and back in the build. See BUG_REVIEW.md (H1).
• Crash hardening: bounds-check encapsulated pixel-data fragment lengths before slicing (TransferSyntaxConverter), handle empty/dot-only TM values (DICOMTime), tolerate duplicate tags inside a sequence item (SequenceItem), and guard a double-resume() race in the Storage SCP association handshake (StorageSCP). See BUG_REVIEW.md (H2, H3, M3, H4).
• Correctness: allowMissingVR in the DICOM JSON decoder had its condition inverted and never inferred a VR from the dictionary; AT-valued elements were byte-swapped as a single 32-bit word instead of two ordered 16-bit words on cross-endian transcode (also added missing OD/OL/SV/UV to the numeric-VR swap set); the data element dictionary loader dropped rows with an empty Name field ((0018,0061), (0400,0315), (300A,0782)). See BUG_REVIEW.md (M1, M2, M4).
• Hardening: segmentation palette index underflow on segmentNumber == 0, slice-unsafe absolute-index reads in ByteOrder/PaletteColorLUT (now relative to startIndex), non-ASCII digits accepted in UID validation, and a 3-byte G1 escape sequence unrecognized at end-of-buffer. See BUG_REVIEW.md (L1-L4, M5).
• CLI / DICOMStudio Workshop parity: rejected negative --frame/--retry/ --parallel/--batch values that previously reached a trapping range or stride instead of erroring cleanly; dicom-anon and its app equivalent now require --output (or --dry-run) instead of silently doing nothing; directory converts in dicom-convert now retag non-DICOM output files with the correct extension per file (previously only the single-file path did); reversed --select ranges in dicom-qr no longer trap; study-level C-GET with --hierarchical now recovers the series UID from the received dataset instead of collapsing to a flat layout; dicom-split now reports real per-file failure counts and exits non-zero when any file failed. See BUG_REVIEW.md ("CLI / DICOMStudio Workshop hardening").

[2.2.10] - 2026-07-15

Added — J2K GPU/CPU Encode Route Planner

• Introduced J2KRoutePlanner, which resolves the three previously conflated encode choices — backend (CPU/GPU/auto), intent (lossy/lossless/lossless-only), and compression type (JPEG 2000 Part-1/Part-2/HTJ2K) — into a single deterministic J2KSwift API call instead of reverse-engineering them from the transfer-syntax UID. auto now genuinely selects the Metal GPU backend for lossy JPEG 2000/HTJ2K encodes where previously it never did. See J2K_ROUTING_ARCHITECTURE.md.
• JPEG 2000 Part-2 (.92/.93) encoding is explicitly rejected with a clear error — the pinned J2KSwift v11.0.2 cannot decode the Part-2 codestreams it encodes — while decoding existing Part-2 files remains fully supported; the real-Part-2 encode path is gated behind a flag for when the library gains decode support.
• Updated CodecBackend/CompressionManager/CompressionConsole wiring and added J2KRoutePlannerTests and J2KGPUEncodeRoundTripTests covering the new routing decisions and GPU round-trip correctness.

Added — CharLS (dcmdjpls) JPEG-LS Bench Peer, repro12bit Repro Tool

• Added CharLSCLICodec (macOS-only), a decode-only DICOMStudio bench peer that wraps DCMTK's dcmdjpls to cross-validate JLSwift-produced JPEG-LS codestreams against a CharLS-backed reference decoder, matching the existing binaryPath/version/decodeFrame surface used by the other CLI peers (djpeg/djxl/Kakadu/Grok).
• J2KTestBenchModels/J2KTestBenchService/J2KTestBenchViewModel/J2KTestBenchView: wired .charls through the JPEG-LS bench family (includeCharLS), and J2KBenchSyntax.all is now derived entirely from TransferSyntax.selectableEncodings for every format (previously only JPEG 2000/HTJ2K rows were catalog-driven), excluding JPEG XL JPEG Recompression (.111) since the bench encodes raw frames rather than repacking an existing JPEG.
• Added repro12bit, a standalone executable target for reproducing/isolating 12-bit codec issues against J2KSwift, alongside JLISWIFT_GAP_ANALYSIS.md documenting a verified bit-depth, color/sampling, and gap audit of the JLISwift dependency (no critical/high findings).
• CompressionQuality.expectedMinPSNRDb gives each encode preset (.low/.medium/.high/ .maximum, and .custom via interpolation) a conservative minimum-PSNR pass bar for lossy round-trip tests, so the bench's lossyPSNRThresholdDb default now tracks J2KTestBenchService.lossyEncodeQuality instead of a fixed 40 dB value a lower-quality preset could never clear.
• ModalityMapping.StandardModality/allCodes centralizes the canonical DICOM modality list; the CLI Workshop's modality pickers (C-FIND/C-MOVE/MWL/etc.) now draw their allowedValues from it instead of hand-maintained arrays, so new modalities need updating in one place.

Fixed — Same-syntax lossy recompression silently dropped --quality

• CompressionManager.isRecompression/compressData/compressDataWithMetrics now treat a same-UID lossy target with an explicit quality as a genuine recompression (decode-to-native
  • re-encode) rather than a byte passthrough. Previously, re-compressing an already-JPEG-2000 (or other lossy-encapsulated) file into the same transfer syntax UID with a new --quality silently copied the existing codestream through unchanged (input size == output size, ~0 ms), discarding the requested quality. Lossless / no-quality same-syntax targets keep the passthrough. dicom-compress and the DICOMStudio Workshop both pass quality through to isRecompression so their two-phase-recompression UI note stays accurate.

Fixed — Standalone macOS packaging

• Removed the DICOMStudio-only OpenJPEG comparison wrapper from DICOMCore's default dependency graph. Standalone macOS consumers no longer require a Homebrew installation or the absolute /opt/homebrew/lib/libopenjp2.a path.
• Removed the corresponding DICOMStudio static-link and arm64-only Xcode settings; production JPEG 2000 support continues to use J2KSwift.
• Optional real-image codec and benchmark tests now report as skipped when the gitignored LocalDatasets or SampleStudies corpora are unavailable in CI.

Changed — J2KSwift v11.0.2

• Updated the JPEG 2000 / HTJ2K / JP3D dependency floor to J2KSwift v11.0.2.
• The decoder-only update stops truncated quality-layer decoding at the exact coding-pass boundary and bounds scratch clearing to the active code-block region, without changing the DICOMKit public API or codestream format.

Added — JPEG Baseline Encoder Choice (DICOMStudio-only)

• Added JPEGCodecEngine (DICOMCore: .jli / .native) and CodecRegistry.encoder(for:engine:), which honours the engine selection only for JPEG Baseline (1.2.840.10008.1.2.4.50) — the one transfer syntax this library can encode two ways: the pure-Swift JLICodec (the registry default for all four JPEG syntaxes) or Apple's NativeJPEGCodec (ImageIO). JPEG Extended/Lossless/Lossless SV1 have no second encoder, so engine is a no-op for them.
• Threaded jpegEngine through CompressionManager.compressData / compressDataWithMetrics, defaulting to .jli so dicom-compress output is unchanged.
• Added an internal "JPEG Engine" picker to the DICOMStudio CLI Workshop's dicom-compress form, visible only for operation == compress && codec == jpeg — a benchmarking aid with no dicom-compress CLI counterpart, so it never appears in the copy-pasteable command preview. Required a new CLIParameterDefinition.visibleWhenAll ([[CLIParameterVisibilityCondition]], ANDed with the existing single-condition visibleWhen) since codec persists across operations and a single condition couldn't pin the picker to compress-only. CommandBuilderHelpers.isVisible(...) is now the one predicate shared by command-preview generation, required-field validation, and the ViewModel's form rendering.

Fixed — dicom-convert to DEFLATE dropped pixel data from encapsulated sources

• An encapsulated source (JPEG/JPEG 2000/RLE/…) converted to Deflated Explicit VR Little Endian is now decoded to native pixels first. DEFLATE (PS3.5 A.5) is a data-set-level codec over a native stream — it has no encapsulated form — but DICOMConverter used to serialize the encapsulated (7FE0,0010) straight into the deflate stream: the codestream survived, but the output was labelled 1.2.840.10008.1.2.1.99 while still carrying an undefined-length, Item-fragmented pixel element, so no conformant reader (including DICOMKit's own) could decode pixel data from it — and the tool reported success.

Fixed — dicom-compress/dicom-convert --output naming a directory failed with "Is a directory"

• Both CLIs now resolve --output through the existing OutputPathResolver (already used elsewhere) before writing, so passing a directory (e.g. ~/Desktop/DICOM_Output/, or whatever the DICOMStudio Workshop's Browse button hands back) writes the input's filename into that directory instead of failing. An explicit file path is still used verbatim.

Fixed — DICOMStudio CLI Workshop could build/compare against a different checkout's DICOMKit

• CLIToolBuilder.repoRoot() and CLIToolTerminalCompare.locateBinary() no longer fall back to a hard-coded absolute path or the process's working directory (which is / for a GUI app). They now resolve the SwiftPM package root by walking up from #filePath — the checkout the running app was actually compiled from — so swift build/binary lookup can no longer silently target a sibling repo whose DICOMKit accepts different tokens, which previously made "Compare CLI" report diffs that didn't exist in the current repo.

Added — Transfer-Syntax Lossy/Lossless Split and Encode-Intent Threading

• Introduced LosslessCapability (losslessOnly/lossyOnly/both), EncodingIntent, and SelectableEncoding on TransferSyntax to correctly model the JPEG 2000/HTJ2K/JPEG XL "general" UIDs (.91/.93/.203/.112), which per PS3.5 may carry either a lossy or lossless codestream. TransferSyntax.parse() itself remains conservative (bare …-lossless still maps to the old reversible-only UID, to avoid changing association negotiation behavior in dicom-send/retrieve/qr) — the lossy/lossless split is exposed only through the new parseEncoding() API. See J2K_HTJ2K_TRANSFER_SYNTAX_SPLIT_PLAN.md.
• Added missing UIDDictionary entries and corrected display names for .92/.93/.201/ .202/.203 (e.g. "JPEG 2000 Lossless Only", "HTJ2K Lossless Only (RPCL)").
• Threaded EncodingIntent from CLI/app codec-name resolution through to JXLCodec and DICOMConverter, so …-lossless codec names now genuinely encode reversibly into the general UID (previously not possible) and …-lossy produces true irreversible compression. See J2K_HTJ2K_JXL_ENCODE_INTENT_AND_LOSSY_ATTRS_PLAN.md.
• Added CompressionManager.applyLossyImageCompressionAttributes, shared by dicom-compress and dicom-convert, which stamps Lossy Image Compression (0028,2110), Method (0028,2114), and Ratio (0028,2112) plus Image Type → DERIVED per PS3.3 C.7.6.1.1.5, with append/once-lossy-always-lossy semantics.
• JXLCodec: added signed 16-bit support (JXLSwift int16 level-shift) and genuine lossy VarDCT grayscale encoding (JXLSwift 1.4.0) instead of silently falling back to lossless.

[2.2.9] - 2026-07-15

Fixed — SwiftPM consumer build hygiene

• Declared the dicom-3d and dicom-j2k target README files as excluded package inputs, removing the two unhandled-file warnings emitted during consumer builds.
• Declared the intentionally inactive complements of the explicitly sourced DICOMKitTests and DICOMViewerTests targets as excluded inputs, removing two package-planning warnings covering 102 test files without changing test membership.
• Applied the existing optional-fixture contract to two HTJ2K benchmark tests so clean CI checkouts skip them when the gitignored MR corpus is unavailable.
• Hardened release validation to reject production compiler warnings and test package-planning warnings, synchronized the release artifact matrix with all 36 enabled CLI products, pinned manual release jobs to the requested tag, and staged release creation as a draft before publication for immutable releases.
• No decoder, runtime, product API, ABI, or command behavior changed.

[2.2.6] - 2026-07-09

Patch release: shared transfer-syntax negotiation token list for dicom-retrieve/dicom-qr, plus two reporting-accuracy fixes (--backend, dicom-compress info lossless state). No association-negotiation or encode-behavior changes — this release only corrects what tools report and what token lists they offer.

Added — Shared transfer-syntax negotiation token list

• Added TransferSyntax.negotiableImageSyntaxTokens / negotiableImageTokens (DICOMCore) as the single source of truth for the transfer-syntax lists offered by the UID-only negotiation tools (dicom-retrieve, dicom-qr) — the negotiation analogue of CompressionManager.supportedCodecs() (dicom-compress) and DICOMConverter.cliTokens (dicom-convert). Every token round-trips through TransferSyntax.parse() (enforced by tests).
• Wired the DICOMStudio CLI Workshop dicom-retrieve / dicom-qr transfer-syntax pickers and the dicom-retrieve / dicom-qr CLI --transfer-syntax help onto that shared list, so both surfaces stay in lockstep. This adds the previously-missing JPEG-LS, JPEG XL, and JPEG 2000 Part 2 (plus explicit-vr-be, deflate, jpeg-extended, jpeg-lossless-sv1) syntaxes that the old hand-maintained lists stopped short of; regenerated the CLIContracts.json parity golden to match.
• CLI Workshop: the dicom-convert transfer-syntax dropdown now shows the short kebab aliases (DICOMConverter.aliasTokens, e.g. jpeg2000-lossless, htj2k-lossy) instead of the CamelCase cliTokens, so it reads the same as the dicom-compress / dicom-retrieve / dicom-qr dropdowns. The dicom-convert CLI resolves the kebab alias identically, so the generated command and in-process execution are unchanged.

Fixed — dicom-compress info reports the true lossless state for general J2K/HTJ2K/JXL UIDs

• CompressionManager.getCompressionInfo now derives isLossless (and the transfer-syntax display name) from Lossy Image Compression (0028,2110) for the both-capable general UIDs (.91/.93/.203/.112), instead of the UID-level flag which always reported "Lossy". A file reversibly encoded into a general UID (e.g. --codec jpeg2000-lossless → .91) now reports Transfer Syntax: JPEG 2000 Lossless / Lossless: Yes on both dicom-compress info surfaces (text + --json) and in the CLI Workshop, so the name and the Lossless line no longer contradict each other. Single-capability UIDs (e.g. .90, .50) are unchanged — their UID is authoritative. Reference: PS3.3 C.7.6.1.1.5.

Fixed — --backend reporting reflected the hardware probe, not the actual encode path

• Added CodecBackendPreference.effectiveEncodeBackend(isLossless:isJPEG2000Family:) (DICOMCore) and CompressionConsole.compressBackend(codec:preference:) (DICOMKit), which report the backend the encoder will actually dispatch to rather than the best available hardware. J2KSwiftCodec only takes the Metal GPU path for a genuinely lossy JPEG 2000/HTJ2K encode — the lossless GPU path isn't bit-exact on 12/16-bit medical data — so auto/--backend metal on a lossless or non-J2K encode previously reported "Metal (GPU)" in dicom-compress/CLI Workshop verbose output even though the encode ran on the CPU. An explicit --backend metal request that can't use the GPU is now downgraded to the CPU backend with an explanatory note instead of silently misreporting.
• CodecBackend.accelerate.displayName now reports "Accelerate (CPU)" instead of the stale "Accelerate (not available)" on platforms where Apple's Accelerate framework is present but the old J2KAccelerate SIMD-family probe (removed in J2KSwift v11.0.0) no longer exists.

[2.2.1] - 2026-07-06

Patch release: strict-concurrency bridge fix plus package-wide warning cleanup. Both swift build and swift build -Xswiftc -warnings-as-errors pass cleanly; no runtime behavior changes.

Fixed — Swift 6 Strict-Concurrency Task Bridge in CLI Targets

• Fixed the DispatchSemaphore + Task {} async-to-sync bridge pattern used by CLI entry points so Swift 6 strict-concurrency no longer flags data-race sendability risks for waitForTask/runAsync helpers. The bridge now uses Task.result and an explicitly documented manually-synchronized capture (nonisolated(unsafe)), preserving runtime behavior while satisfying the compiler's model.
• Updated affected targets:
  • Sources/dicom-jpip/main.swift
  • Sources/dicom-3d/main.swift
  • Sources/dicom-j2k/main.swift
  • Sources/dicom-viewer/main.swift
• Also cleaned up adjacent strict/warnings-as-errors diagnostics surfaced during validation in shared runtime files (ScriptEngine, DICOMDIRReader/DICOMDIRWriter, ImagePreprocessor, ImageResizer, JP3DVolumeBridge, JP3DVolumeDocument, StudyManager, gateway mapping/converter helpers, and compression manager) without changing intended behavior.

Fixed — Remaining -warnings-as-errors Diagnostics Across the Package

• dicom-3d: never-mutated var locals converted to let (VolumeExport, MPRGenerator); redundant try/try? removed from non-throwing DataSet accessor calls (VolumeData).
• DICOMWeb: redundant await removed from synchronous same-isolation calls (ConformanceStatementGenerator, HTTPConnectionPool, HTTPRequestPipeline); unused guard let url binding replaced with a URL(string:) != nil existence test (DICOMJSONDecoder); unused locals dropped (CompressionMiddleware, HTTPRequestPipeline).
• DICOMStudio: pure static J2K test helpers marked nonisolated — they already executed off the main actor at runtime (J2KTestingViewModel); pipe-drain captures in CLIToolTerminalCompare use the same documented nonisolated(unsafe) + happens-before pattern as the CLI bridge; cine timer callback wrapped in MainActor.assumeIsolated (timer is scheduled on the main run loop); slider Binding setters pass closures instead of non-@Sendable function values (DICOMVolumeViewerView, JP3DComparisonView, JP3DVolumeComparisonView); unused locals removed (CLIWorkshopViewModel, JP3DMPRView).

[2.2.0] - 2026-07-04

CLI parity harness, JPEG XL lossy/recompression, and round-trip regression suite.

Changed — P2/P3 Console-Builder and Dir-Walk Dedup (2026-07-04)

Follow-up to the remediation batch: every remaining hand-duplicated console block and directory walk now lives in one shared builder/gatherer that both the CLI and the Workshop call, so the two surfaces can no longer drift (CLI_TOOLS_SHARED_CORE_VERIFICATION.md → P — duplication).

• New shared console builders (CLI text canonical): AnonConsole (+ shared Anonymizer.parseFlexibleTag), PixelEditConsole, ImageConsole, ExportConsole, UIDConsole, CompressionConsole.infoText/infoJSON/backendsText/backendsJSON, NetworkConsole qr-line additions, UPSConsole (DICOMWeb), and NetworkConsole.mwlCreateDetailBlock (dedups the app's HL7/REST MWL-create branches).
• New shared gatherers: FileGatherer.regularFiles(under:recursive:) (sorted, content-agnostic walk — anon/validate/convert/pdf/export/image on both surfaces) and FrameMerger.gatherInputFiles(from:recursive:) + FrameMerger.isDICOMFile (dicom-merge).
• Real drift found and fixed while hoisting (all previously hiding in un-goldened paths): the app's pixedit executor double-printed the Image line, added an "Edited pixel data:" line and a byte-size suffix the CLI never prints; anon's per-file verbose text and single-file failure behavior differed from the CLI; UPS create-workitem/change-state used app-invented wording instead of the CLI's response text; dicom-merge sorted its inputs in-app but not in the CLI, so merged instance order could differ between surfaces (both now sort).
• Deterministic batch order: all shared walks sort by path, so directory-mode output order is stable run-to-run on both surfaces (previously filesystem enumeration order).
• App-only additions that have no CLI counterpart (sandbox redirect notes, UPS pre-flight guidance and error hints, MWL create banners) are unchanged.

Changed — Three-Axis Shared-Core Remediation Batch (2026-07-04)

Follow-up to the three-axis verification of all 40 tools (CLI_TOOLS_SHARED_CORE_VERIFICATION.md, full per-claim outcomes in its Remediation outcomes section). Eleven tools were held by user triage (measure, viewer, 3d, ai, report, gateway, cloud, server, j2k, jpip, print) and the network port/hostname workflow was left untouched as planned. Highlights:

• New shared surfaces (CLI + Workshop call one implementation):
  • DataExchangeWorkflow (DICOMWeb) — the entire dicom-json/dicom-xml pipeline (default output path, always-write-file behavior, tag filtering, metadata-only, verbose lines). Fixes the app's divergent print-to-console/refuse-reverse behavior.
  • ConvertConsole (DICOMKit) — dicom-convert's terse console (transcode line, batch ✓/✗ + Conversion complete: summary); the app's extra Read/Wrote/Transfer-Syntax chrome removed.
  • TagEditConsole (DICOMKit) — dicom-tags change block + Output written to: completion line (the app previously printed an unconditional count and Saved:).
  • QRSessionState (DICOMNetwork) — dicom-qr's save-state model hoisted from the executable; --save-state now also works in the Workshop and app-saved states resume via dicom-qr resume.
  • App validate now renders via the shared ValidationReport (the drifted app copy with its Exit code: annotation block was deleted); app compress batch uses the shared findDICOMFiles; app uid/pixedit call the shared validateFileUIDs/parseRegion.
• Formerly-inert flags implemented: dicom-merge --format enhanced-ct/mr/xa (Enhanced SOP Class + Shared/Per-frame Functional Groups), dicom-script run --parallel (concurrent pipeline commands with source-order output replay), dicom-compress --backend (CompressionConfiguration.forcedBackend → J2K/HTJ2K Metal GPU encode; other codecs unaffected), dicom-image --use-exif under --split-pages (per-page EXIF), dicom-dcmdir update (real implementation via DICOMDIRWorkflow.updateDirectory), dicom-query --referring-physician (now a study-level C-FIND matching key on both surfaces).
• Removed (declared no-ops): dicom-json --format standard|dicomweb and --stream (goldens proved byte-identical output for every value; the encoder always emits the DICOMweb PS3.18 JSON model); the app-only qido timeout field.
• Wire/exit-code correctness: app mpps update no longer sends a studyInstanceUID the CLI never sets; anon in-app exit code is now structured (any failed file → 1) instead of sniffed from output text; dump's No-Color toggle is honored (defaults ON in-app with a truthful --no-color preview).
• Surface adds: uid regenerate accepts multiple inputs in-app (cross-file UID mapping like the CLI); ups gains the CLI's --create <json-file> form; qido gains --verbose.
• New round-trip oracles: merge enhanced-format (SOP class + functional groups + standard unchanged), script parallel-pipeline (rendezvous concurrency + source-order replay + failure propagation), dcmdir update (add + prune-missing).

Added — JPEG XL Lossy Encode (Transfer Syntax …4.112) in dicom-compress

• dicom-compress compress --codec jpeg-xl now produces lossy JPEG XL (1.2.840.10008.1.2.4.112, general JPEG XL) via JXLSwift's VarDCT encoder, alongside the existing lossless JPEG XL (…4.110). Both the dicom-compress CLI and the DICOMStudio CLI Workshop drive the identical shared CompressionManager / CodecRegistry path, so the two surfaces cannot drift. --quality maximum|high|medium|low|0.0–1.0 maps to the JPEG XL quality/distance curve (mirroring the --quality mapping already used by the JPEG and JPEG-LS lossy codecs).
  • Codec naming aligns with the rest of the family (jpeg2000/htj2k): the bare jpeg-xl/jxl names now resolve to the lossy syntax (…4.112); the explicit jpeg-xl-lossless/jxl-lossless names select the lossless syntax (…4.110). jpeg-xl-lossy/jxl-lossy are accepted aliases for the lossy target. (Behaviour change: jpeg-xl/jxl previously encoded lossless.)
  • JXLCodec (Sources/DICOMCore/JXLCodec.swift): added …4.112 to supportedEncodingTransferSyntaxes, a per-instance encodingTransferSyntaxUID that selects the encode mode (…4.110 → lossless Modular, distance 0; …4.112 → lossy VarDCT at a quality-derived distance), and a jxlQuality(from:) mapping. JXLSwift's VarDCT lossy encoder covers 8-/16-bit RGB/RGBA within its size limits; for inputs it can't take (grayscale, oversized) it transparently falls back to the lossless Modular path, so a …4.112 encode always yields a valid — and conformant, since the general syntax permits both — JPEG XL codestream.
  • CodecRegistry (Sources/DICOMCore/ImageCodec.swift): the JPEG XL encoder is now wired per-UID (JXLCodec(encodingTransferSyntaxUID: uid) for …4.110 and …4.112), mirroring the per-syntax encoder wiring already used for JLISwift / J2KSwift.
  • CompressionManager (Sources/DICOMKit/Compression/CompressionManager.swift): codec-name map splits JPEG XL into a lossless entry (jpeg-xl-lossless/jxl-lossless → …4.110) and a lossy entry (jpeg-xl/jxl/jpeg-xl-lossy/jxl-lossy → …4.112). Flows automatically to the CLI validator, the --help codec list, and the DICOMStudio Workshop codec picker (which derives from supportedCodecs()).
  • DICOMStudio (Sources/DICOMStudio/Views/DataExchangeView.swift): the Data Exchange compression-algorithm picker gains "JPEG XL Lossless" and "JPEG XL Lossy" entries, with jpeg-xl marked lossy.
  • Tests (Tests/DICOMCoreTests/JXLCodecRegistryTests.swift, CompressionCodecMapTests.swift): registry now exposes an encoder for …4.112; new round-trips prove an RGB lossy encode decodes back to source dimensions and a grayscale lossy request falls back to lossless (bit-exact); the codec-map parity test pins jpeg-xl → …4.112 and jpeg-xl-lossless → …4.110.
  • Note: dicom-convert's jpeg-xl/jxl --transfer-syntax aliases remain lossless (…4.112 is not yet a convert target) — this change is scoped to dicom-compress.

Added — JPEG XL JPEG Recompression (Transfer Syntax …4.111) in dicom-convert

• dicom-convert --transfer-syntax JPEGXLRecompression now losslessly transcodes a JPEG Baseline (…4.50) file to JPEG XL JPEG Recompression (1.2.840.10008.1.2.4.111) and back. Unlike every other codec, recompression is not a pixel operation: it wraps the existing JPEG bitstream in a JPEG XL container (with a jbrd reconstruction box) so the original JPEG is recovered byte-for-byte — no additional loss on top of the JPEG the file already carries. The reverse (…4.111 → JPEGBaseline) reconstructs the byte-identical original JPEG.
  • JXLCodec (Sources/DICOMCore/JXLCodec.swift): added the fragment-level recompressJPEGFragment(_:) / reconstructJPEGFragment(_:) helpers (backed by JXLSwift's encodeLosslessJPEG / decodeLosslessJPEG), added …4.111 to the decodable transfer syntaxes, and routed the …4.111 pixel decode through JPEG reconstruction → the JPEG codec so decoded pixels match the wrapped JPEG with the descriptor's channel layout (rather than the VarDCT bridge's colour-space representation). The forward encode is deliberately NOT registered as a pixel ImageEncoder (recompression takes JPEG bytes, not pixels).
  • TransferSyntaxConverter (Sources/DICOMCore/TransferSyntaxConverter.swift): added isJXLRecompressionForward / isJXLRecompressionReverse predicates (gated to JPEG Baseline ↔ …4.111, matching JXLSwift's baseline-DCT scope), a fragment-level transcodeJXLRecompression(…) path (no pixel decode/re-encode — structurally mirrors the J2K↔HTJ2K fast path), the canTranscode admission for these pairs, and a lossless-guard bypass so a lossy-Baseline → …4.111 wrap is correctly treated as lossless (it adds no loss).
  • DICOMConverter (Sources/DICOMKit/DICOMConverter.swift): added JPEGXLRecompression to the shared convert target catalog, so the new target flows automatically to the CLI --transfer-syntax help, the DICOMStudio CLI Workshop picker, the representative parameter catalog, and parseTarget(_:) — the CLI and the Workshop share one code path. It is convert-only (not a dicom-compress/CompressionManager pixel codec).
  • CLI parity: CLIContracts.json transfer-syntax abstract updated; cli-parity-gen gains a derived syn-ct-baseline.dcm JPEG-Baseline fixture (ctbaseline) and a dicom-convert ts-JPEGXLRecompression decoded-pixel-hash scenario exercising the shared path in both surfaces.
  • Tests (Tests/DICOMRoundTripTest/ConvertRoundTripTests.swift): oracle round-trips proving byte-identical JPEG reconstruction through the …4.111 wrap, …4.111 pixel decode equals the wrapped JPEG's pixels, onward transcode …4.111 → J2K Lossless, and rejection of a non-JPEG source.

Added — Shared CLI ↔ App Orchestration: CompressionConsole, DICOMConverter, DICOMDIRDumpFormatter, DICOMDIRWorkflow, EncapsulatedDocumentWorkflow

• CompressionConsole (Sources/DICOMKit/Compression/CompressionConsole.swift): Pure shared formatter and input-parser for dicom-compress. Both the CLI binary and DICOMStudio's CLI Workshop call this single type for --quality / --backend parsing, binary byte formatting, and every console line (compress header, stats, batch totals). Replaces duplicated inline formatting on both sides. Mirrors the NetworkConsole shared-formatter pattern.
• DICOMConverter (Sources/DICOMKit/DICOMConverter.swift): Single source of truth for the dicom-convert transfer-syntax conversion API — the ordered target catalog (UID, CamelCase CLI tokens, kebab aliases), parseTarget(_:), cliTokens/aliasTokens picker lists, and the shared convertToDICOM(dicomFile:to:stripPrivate:) pipeline. Both the CLI (dicom-convert) and the DICOMStudio CLI Workshop call this directly so conversion bytes are identical and the --transfer-syntax help text cannot drift from the picker.
• DICOMDIRDumpFormatter (Sources/DICOMKit/DICOMDIRDumpFormatter.swift): Shared renderer for dicom-dcmdir dump output (tree / json / text formats). The CLI and the CLI Workshop both call DICOMDIRDumpFormatter.render(_:format:verbose:) so their output pipelines cannot drift.
• DICOMDIRWorkflow (Sources/DICOMKit/DICOMDIRWorkflow.swift): Shared orchestration helpers for dicom-dcmdir — recursive DICOM file discovery (skipping the existing DICOMDIR, sorting by path), the create build-loop (read, compute relative path, add to DICOMDirectory.Builder, emit summary), and the validate report (statistics + file-set + record-type breakdown). Both the CLI and the Workshop call these helpers, eliminating hand-mirrored orchestration that could silently diverge.
• EncapsulatedDocumentWorkflow (Sources/DICOMKit/EncapsulatedDocument/EncapsulatedDocumentWorkflow.swift): Shared orchestration helpers for dicom-pdf — EncapsulatedDocumentType.fileExtension, defaultModality, the --show-metadata report block, and the human-readable file-size formatter. Both the CLI and the Workshop call these helpers so the dicom-pdf parity surface cannot drift.
• Synthetic DICOMDIR fixture (Sources/DICOMStudio/Resources/CLIParity/synthetic/syn-dicomdir): Pre-built minimal DICOMDIR file added to the synthetic corpus for the offline dicom-dcmdir parity scenario, making the dcmdir dump/validate scenarios runnable without a real PACS file hierarchy.

Fixed — DICOMWriter Drops Encapsulated Pixel Data

• DICOMWriter.serializeElement(_:) now emits encapsulated pixel data (Sources/DICOMCore/DICOMWriter.swift): The generic serializer previously skipped any undefined-length (0xFFFFFFFF) element that was not .SQ, silently emitting an empty PixelData shell. This caused DICOMConverter/dicom-convert to fail on ANY compressed source — the decoder saw zero fragments and threw "Frame 0 starts beyond data bounds" — and would have corrupted any DataSet.write() / DICOMFile.write() rewrite of a compressed dataset. Fixed by adding a dedicated serializeEncapsulatedPixelData(_:fragments:) branch that emits the full Basic Offset Table Item + one Item per codestream fragment (odd fragments padded to even length per PS3.5 A.4) + the Sequence Delimitation Item.

Fixed — DICOMFile.pixelData() No Longer Relabels YBR as RGB

• Removed stale YBR→RGB photometric relabel (Sources/DICOMKit/DICOMFile+PixelData.swift): The previous code relabelled YBR pixel data as RGB for transfer syntaxes that the retired ImageIO-based codecs once handled (JPEG 50/51/57/70, JPEG 2000 90/91). The current registry uses pure-Swift codecs (JLISwift, J2KSwiftCodec, JPEG-LS, RLE) that decode to the source photometric interpretation and leave YBR→RGB conversion to PixelDataRenderer. The relabel suppressed that renderer conversion and washed out/blanked colour compressed previews. Removing it restores correct colour rendering for all compressed sources while preserving the existing isImageIODecodedTransferSyntax helper (unused after this change).

Fixed — JPEG-LS NEAR Parameter Overflow for 16-bit Sources

• JPEG-LS NEAR clamped to 255 (Sources/DICOMCore/JPEGLSCodec.swift): JLSwift's JPEGLSEncoder.Configuration rejects a NEAR value above 255. For a 16-bit source at high quality, the formula maxVal * (1 - quality) * 0.1 could yield NEAR ≈ 655, causing the encode to throw. Added a maxNear = 255 clamp so lossy JPEG-LS encoding of 16-bit images at any quality level succeeds.

Fixed — PixelEditor Handles Compressed Sources

• PixelEditor decodes encapsulated sources before editing (Sources/DICOMKit/PixelEditing/PixelEditor.swift): Editing a compressed (encapsulated) PixelData element in place corrupts the encoded bitstream; the output — still tagged as the compressed transfer syntax — cannot be decoded by a viewer. PixelEditor now detects an encapsulated source, decodes it to native pixels first, and emits the edited result as uncompressed Explicit VR Little Endian. Multi-frame sources are handled correctly.

Fixed — dicom-pixedit --invert Rendered Solid White

• PixelEditor now inverts the VOI window and uses the correct signed pivot (Sources/DICOMKit/PixelEditing/PixelEditor.swift): dicom-pixedit --invert (and the CLI Workshop "Invert" toggle, which shares PixelEditor.processData) produced a solid-white image. applyInvert inverted stored pixels around 2^bitsStored − 1 but never updated the file's VOI Window Center (0028,1050); because the viewer, image exporter, and Horos all honour the stored window by default (DICOMImageExporter.determineWindowSettings, rescale-adjusted), every inverted pixel fell outside the unchanged window and clamped to white. Signed data was additionally clamped because the pivot used the unsigned max instead of −1. Fixes: (1) invert around isSigned ? −1 : maxValue; (2) re-point each Window Center to slope·pivot + 2·intercept − center (output-unit equivalent of inverting the stored center around the pivot), leaving Window Width unchanged and no-op when the file carries no stored window; (3) --apply-window now bakes into — and resets the stored VOI window to — the full representable stored range (signed-aware: [0, 2^b−1] unsigned, [−2^(b−1), 2^(b−1)−1] signed), so a baked signed image is no longer written into only half the range and rendered ~2× too dark. formatDS guards against non-finite values from pathological rescale metadata. Regression tests in PixelEditorTests render the inverted frame through the shared viewer/export window policy and assert a true photographic negative (not solid white) for both MONOCHROME2 and MONOCHROME1, plus the signed window-bake range.

Added — DICOMKitTests: Parity and Regression Test Suite

Seven new test files added to the DICOMKitTests target (Package.swift sources updated):

• EncapsulatedPixelDataWriteTests: Regression for the DICOMWriter encapsulated pixel data fix — asserts the full BOT + fragment structure is emitted, odd fragments are padded, a zero-fragment BOT writes cleanly, and a round-trip DataSet.write() → DICOMParser.parse() recovers the original fragments.
• CompressionManagerImplicitVRTests: Regression for CompressionManager on Implicit VR Little Endian sources — pins that the shared DICOMWriter path re-encodes sequences from parsed items (not raw bytes) and promotes oversized short-VR values to UN, preventing byte-stream desync and the "No pixel data found" failure on decompress.
• CompressedPreviewRenderParityTests: Parity regression that the DICOMFile.pixelData() return value for a freshly compressed file matches the value after a round-trip decompress, for every codec — so the photometric-relabel removal cannot silently reintroduce colour corruption.
• CompressionConsoleTests: Contract tests that lock the exact dicom-compress console strings produced by CompressionConsole — byte formatting, quality parsing, header lines, compressed-result lines — so the CLI and CLI Workshop cannot drift.
• DICOMConverterTests: Contract tests for the shared DICOMConverter API — every catalog token (cliToken / aliasToken / UID / extraAliases) round-trips through parseTarget; picker token lists are exactly the catalog; CLIContracts.json entry is regenerable from the catalog; DEFLATE converts without error.
• ExportWindowParityTests: Regression that DICOMImageExporter.renderFrameForExport and determineWindowSettings rescale-adjust the VOI window (HU → stored via Rescale Slope/Intercept), so a CT with a non-zero Rescale Intercept exports with the correct contrast — not washed out.
• PixelEditorTests: Regression that PixelEditor decodes encapsulated sources to native pixels before editing, emits Explicit VR Little Endian output, and preserves tag edits in the serialized file.

Added — WADORetrieveConsoleFormatter (Shared WADO-RS / WADO-URI Retrieve Renderer)

• WADORetrieveConsoleFormatter (Sources/DICOMWeb/WADORetrieveConsoleFormatter.swift): Shared output renderer for WADO-RS / WADO-URI retrieve — verbose preamble blocks, per-mode status lines (metadata / rendered / thumbnail / frames / instances / WADO-URI result), and the metadata body (JSON pretty-printed + PS3.19 Native DICOM Model XML). Mirrors QIDOResultFormatter (query) and UPSResultFormatter (ups): a SINGLE formatter both sides call, so the dicom-wado retrieve CLI binary and DICOMStudio's in-app retrieve cannot produce different output.
  • DICOMWado.swift (RetrieveCommand) now delegates all verbose preamble, per-mode status, and metadata body output to WADORetrieveConsoleFormatter instead of hand-rolling inline strings.
  • CLIWorkshopViewModel.swift (WADO retrieve case) likewise delegates to the formatter; the mode-detection / inline-echo block is removed, and the verbose preamble is gated by --verbose on both sides identically.
  • parseFrameNumbers moved from RetrieveCommand into WADORetrieveConsoleFormatter (as a throwing method returning [Int]) with a companion WADOFrameParseError type; the CLI catches WADOFrameParseError and re-throws as ValidationError.

Added — STOWResultFormatter (Shared WADO STOW-RS Upload Renderer)

• STOWResultFormatter (Sources/DICOMWeb/STOWResultFormatter.swift): Shared console renderer for dicom-wado store (STOW-RS) upload output — verbose pre-upload header, per-batch start/result lines, per-failure detail, and the always-printed final summary block. Both the dicom-wado store CLI path and DICOMStudio's in-app STOW upload call this single formatter, preventing output pipeline drift. The summary block format is a parity contract that CLIParityWADOComparator.parseStore anchors on.

Added — UPS-RS Parity Harness: Full Operation Matrix, Global Subscribe, and get --format/--verbose

• UPS write scenarios run out-of-the-box: The parity harness no longer gates the full UPS operation matrix on a user-supplied Procedure Step Label. A upsDefaultLabel ("CLI Parity Workitem") is substituted when the WADO panel's label is blank, so ups-lifecycle, ups-lifecycle-complete, ups-lifecycle-cancel, ups-get, ups-create-attrs, ups-create-json, and ups-subscribe always appear in the scenario list — matching how the harness already auto-picks the AE title and station filter.
• Global UPS subscribe scenario (ups-subscribe-global): New runWADOUPSSubscribeGlobalScenario runner exercises ups --subscribe --aet <ae> (no --workitem-uid) → ups --unsubscribe --aet <ae> — the GLOBAL round-trip that subscribes to ALL workitems' events. Reference uses DICOMwebClient.subscribeToAllWorkitems + unsubscribeFromWorkitem(nil). Parity on round-trip outcome; servers that don't enable UPS subscription fail both sides identically (failureAgreement).
• ups-get --format / --verbose variants: Four --format flag variants (table, json, csv) plus a --verbose variant are now generated for the ups-get scenario. The flags are threaded through studioParams["get-format"] and "get-verbose" and appended at run time (after the Workitem UID is known), mirroring how the CLI appends them to the chained ups --get <uid> command.
• Transaction UID flow corrected: The UPS lifecycle runner (runWADOUPSLifecycleScenario) no longer pre-mints a Transaction UID and supplies it to the --update --state IN_PROGRESS claim. Instead it lets the server assign one, parses it from the CLI's IN PROGRESS output (Transaction UID: …), and reuses it for the terminal COMPLETED/CANCELED transition — exactly how a real operator works. The reference (CLIParityNetworkReference.wadoUPSLifecycle) likewise captures and reuses claimResp.transactionUID. When no UID is returned the terminal transition is skipped and recorded as not reached.
• wadoUPSSubscribeGlobal reference method added to CLIParityNetworkReference: calls client.subscribeToAllWorkitems + client.unsubscribeFromWorkitem(nil); createOK is vacuously true (no workitem is created).

Changed — C-GET and dicom-send Dry-Run Comparators Aligned with Shared Formatters

• C-GET comparator (CLIParityRetrieveComparator): The shared NetworkConsole.cGetSummary now emits exactly one terse line — "✅ C-GET completed — N file(s) received" on success or "⚠️ C-GET completed but received 0 instances. …" when nothing arrived — instead of a structured C-GET Completed: block with sub-operation counts. The parser now reads the received-file count from that line only; completed/failed are no longer parsed or compared for C-GET (they are unobservable in the CLI text). canonical() updated accordingly: C-GET compares success + files; C-MOVE still compares completed + failed + warning.
• dicom-send dry-run comparator (CLIParitySendComparator): The shared formatter's dry-run path (NetworkConsole.sendHeader) prints the gathered file count in the header's "Files: N" field rather than "Found N file(s) to send". The parser now reads the first "Files:" line — the header count — rather than "Found".

Changed — UPS CLI Workshop: unsubscribe Operation and Simplified cliMapping

• unsubscribe operation added to the UPS parameter definition in CLIWorkshopHelpers: the operation picker now lists search, get, create-workitem, change-state, subscribe, unsubscribe. --workitem-uid is shown for unsubscribe as well as subscribe and create-workitem.
• --search and --create-workitem moved to cliMapping: Both flags are now emitted automatically when the matching operation tab is selected, removing the separate boolean-toggle CLIParameterDefinition entries that were previously needed. This mirrors the existing --subscribe/--unsubscribe mapping pattern.
• No auto-pre-selection in Network mode: Switching to Network mode no longer pre-selects the first network tool. The user explicitly picks which tools to include in the parity sweep.

Fixed — HL7 ORM^O01 Field Placement for dcm4chee-arc MWL Create

• HL7 ORM IPC segment + OBR field map corrected (ModalityWorklistService.buildHL7ORM): The previous implementation wrote scheduledStationAETitle into OBR-20, which dcm4chee-arc's default inbound order stylesheet (hl7-order2dcm.xsl) reads as the Scheduled Procedure Step ID (0040,0009) — so a user's Station AET surfaced on the server as the SPS ID. Fixed in two ways:
  • OBR path corrected: rebuilt with an explicit index→value map (hl7Segment(_:fields:) helper) so the values land at their exact positions. OBR-18 = Accession Number, OBR-19 = Requested Procedure ID, OBR-20 = SPS ID, OBR-24 = Modality, OBR-27 4th component = SPS Start Date/Time.
  • IPC segment added: a dcm4che-private IPC (Imaging Procedure Control) segment is emitted after OBR so every SPS attribute has an unambiguous, configuration-independent slot — IPC-7 = Station Name, IPC-9 = Scheduled Station AE Title (the only ORM path that carries them). IPC-1/2/3 also supply Accession / Requested Procedure ID / Study Instance UID, matching the OBR fallback exactly.
  • buildHL7ORM promoted from private to internal to allow the new field-placement regression tests (Tests/DICOMStudioTests/MWLCreateHL7ORMTests.swift) to assert each value's exact HL7 position without requiring a live MLLP server.

Fixed — WADO-URI Endpoint Resolution for dcm4chee5

• WADOURIClient.resolveURIEndpoint(_:) (new public static method): dcm4chee-arc 5.x serves WADO-URI (?requestType=WADO) from /wado, while the sibling WADO-RS/QIDO-RS endpoint lives at /rs. Supplying a WADO-RS base URL for a WADO-URI request returned HTTP 404. The resolver rewrites a trailing /rs path segment to /wado; all other base URLs are returned unchanged. Because the dicom-wado CLI, CLI Workshop, and parity reference all retrieve through this one client, they resolve identically and cannot drift.

Fixed — dicom-mpps N-CREATE Status Guard

• dicom-mpps create --status validation: N-CREATE must always start the step IN PROGRESS; the previous code accepted COMPLETED or DISCONTINUED at creation, which servers reject (terminal states are reached only via N-SET). The create subcommand now validates that --status is IN PROGRESS and emits a clear ValidationError directing the user to dicom-mpps update for state transitions.

Added — UPS-RS Result Formatter (Shared)

• UPSResultFormatter (Sources/DICOMWeb/UPSResultFormatter.swift): Shared output renderer for UPS-RS worklist search results — table, JSON (UPSOutputFormat), and CSV — used by both the dicom-wado ups --search CLI path and DICOMStudio's in-app UPS worklist search. Mirrors QIDOResultFormatter (QIDO-RS) and DICOMQueryResultFormatter (DIMSE): a single formatter both sides call so their output pipelines cannot drift.

Added — CLI Workshop PACS Server Edit

• Edit saved PACS server profiles: The CLI Workshop saved-server list now supports in-place editing (beginEditServer(id:) / saveEditedServer() on CLIWorkshopViewModel). A new showEditServerSheet / editingServerID pair drives the edit sheet; saving re-applies the updated values when the edited profile is currently selected. Previously only add and delete were supported.

Changed — NetworkConsole Shared Formatter Covers All Network CLIs

• NetworkConsole (DICOMNetwork) now covers all DIMSE network tools: dicom-echo, dicom-mwl (query), and dicom-mpps joined the shared formatter, completing the set started with dicom-query / dicom-send / dicom-retrieve / dicom-qr / dicom-wado. All human console output — headers, per-echo progress, summaries, verbose details — routes through one NetworkConsole method on both the CLI binary and the DICOMStudio CLI Workshop in-process path, making terminal-compare diff drift impossible by construction.
• dicom-send/ProgressReporter.swift removed: its logic was absorbed into NetworkConsole. Any callers that imported it directly must switch to the corresponding NetworkConsole.* methods.

Added — Network CLI & DICOMweb Tests

• MWLCreateHL7ORMTests (Tests/DICOMStudioTests/): Regression tests asserting each value in the HL7 ORM^O01 message built by ModalityWorklistService.buildHL7ORM lands at its exact field position in both the OBR fallback path and the IPC segment, so the field-placement bug (OBR-20 Station AET mismap) cannot silently return.
• UPSTests (Tests/DICOMWebTests/): Coverage for UPS-RS workitem query parsing and the new UPSResultFormatter output (table/JSON/CSV).
• WADOURIClientTests (Tests/DICOMWebTests/): Coverage for WADOURIClient.resolveURIEndpoint (no-op for /wado, rewrite for /rs, passthrough for other paths) and WADO-URI URL building.

Added — Network Utility (Live Terminal Output)

• Network Utility panel (NetworkUtilityView, NetworkUtilityViewModel, NetworkUtilityService): Six-tab general-purpose network diagnostics tool surfaced as a new sidebar destination in DICOMStudio.
  • Ping — wraps /sbin/ping; live per-packet output streams into a terminal panel, parsed summary (min/avg/max RTT, packet loss) replaces it on completion.
  • Port Scanner — concurrent TCP probes via NWConnection; results append in arrival order for a live scan log, sorted by port number on completion.
  • Traceroute — wraps /usr/sbin/traceroute; each hop line streams as it resolves; stderr merged into stdout so the traceroute to … header appears at the top in real time.
  • DNS Lookup — wraps /usr/bin/dig per selected record type (A, AAAA, MX, TXT, NS, CNAME, SOA, PTR); each query echoes a $ dig … header then streams its answer block.
  • Interfaces — lists all network interfaces with IPv4/IPv6 addresses, MAC address, MTU, flags, and status badges.
  • Netstat — wraps /usr/sbin/netstat; streams TCP/UDP connections or routing table live; parsed counts (listening/established/routes) shown on completion.
• Shared host input: A single sharedHost field is shared by the Ping, Port Scanner, and Traceroute tabs — typing a host in any one of them pre-fills the others.
• AsyncStream<String>-based live streaming (runStreamingProcess): All five process-based tools share a single streaming process runner; stdout and stderr are merged into one pipe so output arrives in natural order, then yielded chunk-by-chunk via AsyncStream.
• UTF-8 carry-over buffer: A var pending = Data() accumulator in the availableData read loop ensures multibyte characters (IDN hostnames, TXT/PTR record content) are never split and silently dropped between reads.
• Run-identity guard (streamGeneration / portScanGeneration): Each run captures a generation counter; onChunk closures and completion assignments check self.streamGeneration == gen and discard stale deliveries from cancelled or superseded runs.
• SIGKILL escalation: Both the wall-clock watchdog and ProcessKillBox.cancel() send SIGTERM then escalate to SIGKILL after a 3-second grace period, preventing hung processes from blocking the UI indefinitely.
• Watchdog liveness guard: The watchdog DispatchWorkItem checks proc.isRunning before acting, preventing a process that exits naturally at the deadline from being mislabelled as timed out.

[2.1.0] - 2026-05-21

DICOMStudio: J2K Test Bench, responsive layout, and imaging-first navigation.

[2.0.0] - 2026-05-21

Added — J2KSwift v3.2.0 Integration (Phases 1–9)

• J2KSwift v3.2.0 codec stack (Sources/DICOMCore/J2KSwiftCodec.swift, HTJ2KCodec.swift, JP3DCodec.swift): Replaces Apple ImageIO as the primary JPEG 2000 path on all platforms, enabling full Linux support via a pure-Swift scalar backend.
  • J2KSwiftCodec: Handles JPEG 2000 Lossless (.90), JPEG 2000 Lossy (.91), Part 2 Lossless (.92), Part 2 Lossy (.93) with 8/12/16-bit grayscale and RGB support.
  • HTJ2KCodec: Full HTJ2K Lossless (.201), HTJ2K RPCL Lossless (.202), HTJ2K Lossy (.203). Fast-path transcoder via J2KTranscoder (no pixel decode); 5.4× decode speedup over J2K on macOS arm64.
  • JP3DCodec: ISO/IEC 15444-10 volumetric encoding/decoding for multi-frame CT/MR/PET series with lossless, lossless-HTJ2K, and lossy modes.
• JPIP streaming (Sources/DICOMKit/DICOMJPIPClient.swift): Progressive 2D and 3D tile streaming for large remote studies; transfer syntaxes JPIP Referenced (.94) and JPIP Referenced Deflate (.95) registered.
  • dicom-jpip CLI tool with fetch, uri, serve, and info subcommands.
  • DICOMFile.openVolumeProgressively(serverURL:sliceJPIPURIs:qualityLayers:) API for huge CT/MR datasets.
• JP3D volume bridge (Sources/DICOMKit/JP3DVolumeBridge.swift): Converts multi-frame DICOM series ↔ J2KVolume; preserves SliceLocation, ImagePositionPatient, SeriesInstanceUID.
  • JP3DVolumeDocument: Encapsulated document SOP (private SOP 1.2.826.0.1.3680043.10.511.10) with .jp3d payload + JSON sidecar; MIME type application/x-jp3d.
  • DICOMFile.openVolume(from:) / openVolume(from:jpipServerURL:) for unified volume access.
• Hardware acceleration (CodecBackend enum): Metal (Apple GPU), Accelerate (SIMD), scalar fallback; CodecBackendProbe selects best available at runtime. --backend flag on dicom-compress and dicom-3d.
• dicom-j2k CLI tool (8 subcommands): info, validate, transcode, reduce, roi, benchmark, compare, completions. 53 tests.
• DICOMStudio enhancements:
  • Progressive decoding with ProgressiveDecodeModel / ProgressiveImageView (AsyncStream-driven .quarter → .half → .complete state machine).
  • ROI decoding wired to pinch-zoom gestures.
  • JP3D MPR views (axial / sagittal / coronal) via JP3DMPRViewModel / JP3DMPRView.
  • JPIP loader with quality-layer slider.
• Transfer syntaxes added to registry, DICOMValidator, and StorageSCP presentation contexts: .htj2kLossless, .htj2kRPCLLossless, .htj2kLossy, .jpip, .jpipDeflate, .jpeg2000Part2Lossless, .jpeg2000Part2.
• DICOMweb HTJ2K media types: image/jph and image/jphc advertised in capability; WADO-RS accept headers updated.
• dicom-compress, dicom-convert, dicom-send, dicom-retrieve, dicom-viewer, dicom-info, dicom-validate extended for HTJ2K, JP3D, and JPIP transfer syntaxes.
• Codec Inspector panel in DICOMStudio: shows decoder name, backend (Metal/Accelerate/scalar), and decode timing.

Fixed

• JPEG 2000 16-bit rendering pipeline: Fixed near-black output after conversion when preserving original bit depth

  • Normalized ImageIO-decoded 16-bit JPEG 2000 samples back to the DICOM Bits Stored range in NativeJPEG2000Codec
  • Preserved original metadata for JPEG 2000 and JPEG 2000 Lossless conversions (BitsAllocated, BitsStored, HighBit)
  • Verified CT-style datasets with VOI/Rescale tags render correctly after implicit VR → JPEG 2000 lossless transcoding

• DICOM Studio metadata consistency: Fixed transfer syntax source ordering in metadata loading

  • MetadataViewModel now prefers File Meta Information (0002,0010) before dataset fallback
  • Aligns metadata display behavior with converted-file transfer syntax as stored on disk

• Test Infrastructure: Fixed platform-specific test compilation errors

  • Added #if canImport(CoreGraphics) guards to ColorTransformTests for Apple platform-only APIs
  • Fixed DataElement initializer calls in ICCProfileAdvancedTests with missing length parameters
  • Fixed ambiguous type references in SegmentationParserTests
  • Tests now compile cleanly on Linux CI runners and Apple platforms

Changed - DICOM Standard Edition Update

• Updated DICOM standard reference from 2025e to 2026a
  • The 2026a release is now the current edition available at https://www.dicomstandard.org/current/
  • Updated dicomStandardEdition constant to "2026a"
  • Updated all source code doc comments referencing DICOM PS3.x editions
  • Updated conformance statement, FAQ, contributing guide, and README
  • Key differences from 2025e to 2026a:
    • New supplements including CT Image Storage for Processing (Sup252)
    • Radiation Dose Structured Report (RDSR) informative annex (Sup245)
    • Enhanced DICOMweb services (Sup248, Sup228)
    • Data dictionary and controlled terminology updates
    • Correction proposals addressing encoding clarifications and CID additions
    • Improved sex and gender data representation
    • Frame Deflate transfer syntax enhancements for segmentation encoding

[1.2.6] - 2026-02-07

Added - Phase 5 CLI Tools Complete

• dicom-mpps (v1.2.6): Modality Performed Procedure Step (MPPS) operations
  • N-CREATE operation for creating MPPS instances (procedure start)
  • N-SET operation for updating MPPS instances (procedure completion/discontinuation)
  • Support for IN PROGRESS, COMPLETED, and DISCONTINUED states
  • Referenced SOP instance tracking
  • MPPSService in DICOMNetwork module
  • Complete CLI tool with create and update subcommands
  • Documentation and README

[1.2.5] - 2026-02-07

Added - Phase 5 CLI Tools

• dicom-mwl (v1.2.5): Modality Worklist Management
  • C-FIND query support for Modality Worklist Information Model
  • WorklistQueryKeys with flexible filtering (date, station AET, patient, modality)
  • JSON output support for automation
  • Verbose mode for detailed attribute display
  • ModalityWorklistService in DICOMNetwork module
  • Complete CLI tool with query subcommand
  • Documentation and README

1.0.0 - TBD

Major Release - Production Ready

This is the first production-ready release of DICOMKit, a pure Swift DICOM toolkit for Apple platforms (iOS 17+, macOS 14+, visionOS 1+).

Core Features (v0.1-v0.5)

DICOM File Support

• Reading & Parsing: Full support for reading DICOM files with comprehensive parsing
• Transfer Syntaxes:
  • Explicit VR Little Endian (1.2.840.10008.1.2.1)
  • Implicit VR Little Endian (1.2.840.10008.1.2)
  • Explicit VR Big Endian (1.2.840.10008.1.2.2)
  • Deflated Explicit VR Little Endian (1.2.840.10008.1.2.1.99)
• Data Types: All standard DICOM Value Representations (VR) supported
• Specialized Types: Date, Time, DateTime, AgeString, PersonName, UniqueIdentifier, etc.
• Writing: Create and modify DICOM files with proper serialization
• UID Generation: Utilities for creating unique DICOM identifiers

Pixel Data Support (v0.3-v0.4)

• Uncompressed Images: Support for all standard photometric interpretations
  • MONOCHROME1, MONOCHROME2
  • RGB, PALETTE COLOR
• Compressed Images: Native codec support for:
  • JPEG Baseline (Process 1)
  • JPEG Extended (Process 2 & 4)
  • JPEG Lossless & JPEG Lossless SV1
  • JPEG 2000 (Lossless and Lossy)
  • RLE Lossless (pure Swift implementation)
• Multi-frame Support: Handle image sequences
• CGImage Rendering: Native Apple platform integration for display
• Windowing: Window Center/Width support for grayscale images

Networking Features (v0.6-v0.7)

DICOM Network Protocol (DIMSE)

• Core Infrastructure: PDU handling, association management
• C-ECHO: Verification service for connectivity testing
• C-FIND: Query services for searching DICOM archives (Patient, Study, Series, Image levels)
• C-MOVE & C-GET: Retrieve services for fetching studies and images
• C-STORE: Storage services (SCU and SCP)
  • Single file and batch storage operations
  • Progress tracking with AsyncSequence
  • Storage SCP for receiving files
• Storage Commitment: N-ACTION based commitment verification
• Advanced Features:
  • TLS/SSL support for secure connections
  • Connection pooling and reuse
  • Association timeout configuration
  • Asynchronous API with Swift Concurrency

DICOMweb Services (v0.8)

RESTful Web Services

• WADO-RS: Retrieve studies, series, and instances via HTTP
  • Multi-part response parsing
  • Metadata retrieval
  • Rendered image support
• QIDO-RS: Query services over HTTP
  • Study, series, and instance queries
  • Fuzzy matching support
  • Pagination with limit/offset
• STOW-RS: Store instances via HTTP multipart upload
  • Batch upload support
  • Progress tracking
• UPS-RS: Unified Procedure Step worklist services
  • Workitem creation, retrieval, updates
  • State transitions (SCHEDULED → IN PROGRESS → COMPLETED/CANCELED)
  • Subscription support for notifications
• Authentication: Bearer token and OAuth2 support
• TLS: Secure HTTPS connections with custom certificate validation

Structured Reporting (v0.9)

SR Document Support

• Core Infrastructure: SR IOD parsing and document tree navigation
• Document Types: Support for all standard SR templates
  • Basic Text SR, Enhanced SR, Comprehensive SR
  • Key Object Selection
  • Measurement reports
  • CAD SR (Chest, Mammography)
• Content Items: All relationship types and value types supported
• Coded Terminology: SNOMED CT, LOINC, RadLex integration
• Measurement Extraction: Automated extraction of measurements and coordinates
• Document Creation: SR document builders with template validation
• Template Support: TID 1500 (Measurement Report), TID 1400 (Chest CAD SR), and more

Advanced Features (v1.0.1-v1.0.13)

Presentation States

• Grayscale Presentation State (GSPS): Annotations, LUT transformations, spatial transforms
• Color Presentation State (CSPS): Color management, blending operations
• Pseudo-Color: Color lookup tables, hot/cold mapping

Hanging Protocols

• Protocol Definition: Screen layout and viewport configuration
• Matching Logic: Image set selection based on modality, anatomy, laterality
• Display Sets: Multi-image layout management

Radiation Therapy (RT)

• RT Structure Set: ROI contours, structure visualization, volume calculation
• RT Plan: Beam definitions, treatment machine setup
• RT Dose: Dose grids, isodose curves, DVH (Dose-Volume Histogram)

Segmentation

• SEG IOD: Binary and fractional segmentation support
• Rendering: Segment overlay with configurable colors
• Builder API: Create segmentation objects programmatically

Parametric Maps

• Quantitative Imaging: Float pixel data support
• Real-World Value Mapping: Physical units, SUV calculation
• ICC Color Profiles: Professional color management

International Support

• Character Sets: ISO 2022, ISO 8859, GB18030, GBK, EUC-KR, Shift_JIS, UTF-8
• Private Tags: Vendor-specific tag dictionaries (GE, Siemens, Philips)

Performance

• Memory Optimization: Efficient large file handling
• SIMD Acceleration: Vectorized operations for image processing
• Lazy Loading: On-demand pixel data decompression

Documentation

• DocC Catalogs: Comprehensive API documentation
• Platform Guides: iOS, macOS, visionOS integration guides
• DICOM Conformance: Formal conformance statement

Example Applications (v1.0.14)

DICOMViewer iOS

• Multi-modality image viewer with gesture controls
• Windowing, pan, zoom, measurements
• Hanging protocol support
• Series browser with thumbnails
• Local file import and PACS integration

DICOMViewer macOS

• Removed from the repository.

Command-Line Tools

• dicom-info: Display DICOM file metadata
• dicom-dump: Detailed data element dump
• dicom-convert: Transfer syntax conversion
• dicom-anon: Anonymization tool
• dicom-validate: Conformance validation
• dicom-query: PACS query tool
• dicom-send: DICOM network send utility

Sample Code & Playgrounds

• 27 Xcode Playgrounds demonstrating library features
• Integration examples for iOS, macOS, visionOS
• Network protocol examples
• Image processing examples

Technical Highlights

• Pure Swift: No Objective-C runtime dependencies
• Swift 6 Compliant: Full strict concurrency support
• Platform Native: Leverages Apple frameworks (ImageIO, CoreGraphics, RealityKit)
• Modern API: Swift Concurrency (async/await), Sendable types
• Comprehensive Testing: 1,920+ tests across core, networking, and applications
• Medical Imaging Standards: DICOM PS3.x compliant

Platform Support

• iOS: 17.0 and later
• macOS: 14.0 and later
• visionOS: 1.0 and later
• Swift: 6.0 and later

Dependencies

• Swift Argument Parser 1.3+ (for CLI tools only)

Installation

Swift Package Manager

Add to your Package.swift:

dependencies: [
    .package(url: "https://github.com/Raster-Lab/DICOMKit.git", from: "1.0.0")
]

Documentation

• README.md - Overview and quick start
• CONTRIBUTING.md - Contribution guidelines
• MILESTONES.md - Development roadmap
• Documentation/ - API documentation and guides

Known Limitations

• Network integration tests require access to test PACS systems (documented for future)
• Transfer syntax conversion deferred to future versions
• Some advanced character sets deferred (ISO IR 100 extended)
• Store-and-forward networking features deferred

Security & Privacy

• No known vulnerabilities in dependencies
• HIPAA considerations documented
• PHI (Protected Health Information) handling guidelines provided
• Secure network communication with TLS support

Breaking Changes

This is the first stable release (v1.0.0). Future breaking changes will only occur in major version updates (2.0, 3.0, etc.).

Contributors

Built with ❤️ by the DICOMKit team and contributors.

License

See LICENSE file for details.

---

Pre-release History

For detailed development history of pre-release versions (v0.1 - v0.9, v1.0.1 - v1.0.15), see MILESTONES.md.

Notable Pre-release Versions

• v0.1: Core infrastructure, basic file parsing
• v0.2: Extended transfer syntax support
• v0.3: Pixel data access
• v0.4: Compressed pixel data
• v0.5: DICOM writing
• v0.6: Networking (Query/Retrieve)
• v0.7: Storage services
• v0.8: DICOMweb
• v0.9: Structured Reporting
• v1.0.1-v1.0.13: Advanced features
• v1.0.14: Example applications
• v1.0.15: Production release preparation

---