Version 1.4#
DataLab Version 1.4.1#
Requirements#
Runtime requirements are unchanged since version 1.4.0.
Third-party:
PyYAML is now part of the
testoptional dependencies: the test suite needs it to check the projects created by the plugin generator.
🛠️ Bug Fixes since version 1.4.0#
Tutorial examples and metadata: The custom denoising example no longer produces undefined values on constant image regions, and the Spyder tutorial no longer produces undefined values on grids containing the first peak position. The metadata paste dialog now explains that annotations are copied and pasted separately (fixes Issue #567).
Scientific processing (⚠️ results change): With Sigima 1.4.1 or later, X–Y mode uses only the signals’ common interval instead of extending one signal beyond its data, while retaining the interpolation confirmation; disjoint domains are refused. The enclosing circle now contains the selected pixel centers in physical coordinates, quadratic interpolation passes through the samples, pulse offsets retain their sign, and flat-field correction refuses zero denominators on corrected pixels. Previously recorded calls of these operations are refused at replay rather than silently giving different results; existing stored data is unchanged (fixes Issue #565 and Sigima Issue #130).
Integer images and X–Y mode (⚠️ results change): with Sigima 1.4.1 or later, on signed integer images whose range of values exceeds the capacity of their type (e.g.
int16values from -30000 to +30000), Analysis > Minimum enclosing circle center no longer gives a much too large circle, Analysis > Contour detection finds the contours, and Analysis > 2D peak detection no longer reports spurious peaks. Flat-field correction of an integer image now gives a floating-point image instead of wrapping the values exceeding the range of the image type. With a region of interest on the first signal, X-Y mode and Replace X by other signal’s Y now keep only the points within this region, instead of mixing the original coordinates of the other points into the result. X-Y mode results now take the titles and units of the Y axes of the two signals, signals with different X units are refused, and signals sharing the same X grid are paired without interpolation. Previously recorded calls of the changed operations are refused at replay (fixes Issue #571 and Sigima Issue #136).Plugin installation: installing a plugin whose required packages are missing now names these packages and tells how to provide them, instead of reporting an unexplained “Dependency wheelhouse not found” error with a temporary folder.
Brightness and contrast: the Minimum and Maximum fields of Processing > Exposure > Brightness and contrast (in its dialog and in the Processing tab of the properties panel) showed numbers with 17 significant digits in fields too narrow for them, so that their beginning was cut off, e.g.
835736228608e-43instead of6.5127532828723551e-44. These fields now show the complete value, with a precision matching that of the sliders (e.g.0and1.7557), and the exact value in their tooltip; a value which is not modified is kept exactly (fixes Issue #510).Processing parameters: moving the slider of a numeric parameter gave values such as
0.30000000000000004or1.2000000000000002instead of0.3or1.2, which were then used and recorded as processing parameters, e.g. with Sigmoid correction and Adaptive histogram equalization (Processing > Exposure), the Butterworth filter (Processing > Frequency filters) or the signal Peak detection (Analysis). Slider values are now free of these spurious digits (fixes Issue #510).Image coordinates: since DataLab 1.2.0, Set uniform coordinates was missing from Processing > Axis transformation in the Image panel, so this feature could no longer be run from the menus. It is available again, next to Polynomial calibration (fixes Issue #515).
Web API: starting the Web API server through the remote control (XML-RPC), e.g. with
RemoteProxy.start_webapi_server(), used port 8080 instead of the automatic port (18080, or the next free one), so that DataLab-Kernel could not find the server automatically. Unless a port is given, the server now uses the automatic port, as when it is started from File > Web API (fixes Issue #516).AI Assistant: the Enable AI Assistant option of the settings had no effect: the assistant remained available, although the documentation stated that no request was sent when it was disabled. This option is removed: the assistant contacts its provider only when you send a message or click Test connection, as the documentation now states. A new Settings… button of the AI Assistant panel opens the AI Assistant tab of the settings, and the messages of the panel point to File > Settings instead of a nonexistent Edit > Preferences menu (fixes Issue #571).
Visualization settings: the Downsampling max points field could remain editable when the settings dialog opened with Use auto downsampling unchecked, e.g. after checking this option and cancelling the dialog. It now follows the checkbox as soon as the dialog opens (fixes Issue #571).
Stability analysis: Overlapping Allan variance was missing from Processing > Stability analysis in the Signal panel, so it could only be computed with the other methods, through All stability features…. It is now listed with the other variances (fixes Issue #518).
Gaussian bandpass filter: the Inverse FFT result parameter of Processing > Frequency filters > Gaussian bandpass had no effect: Absolute value gave the same image as Real part, although the parameter was recorded in the provenance as if it had been applied. With Sigima 1.4.1 or later, Absolute value gives the modulus of the inverse FFT, and the title of the result shows the chosen option; Real part, the default, gives the same result as before. Steps of this filter recorded with an earlier version are refused when they are verified or replayed, instead of giving another result (fixes Sigima Issue #123).
Cardinal sine signal: the cardinal sine of Create > Cardinal sine, described as
y = y0 + A sinc(2π f x + φ), used the normalized definitionsinc(u) = sin(πu)/(πu): with a frequency of 1 Hz, its first zeros were at about ±0.159 instead of ±0.5. With Sigima 1.4.1 or later, it uses the definitionsinc(u) = sin(u)/u(withsinc(0) = 1), as the 2D sinc image, which is unchanged; the creation dialogs of both now show this definition. For the same parameters, a cardinal sine signal created, edited or recomputed from the History panel is now different; signals already created, e.g. in a saved workspace, keep their data (fixes Sigima Issue #60).Web API: when required components could not be imported, the menu and remote control suggested installing a nonexistent extra. They now name the required dependencies and explain how to check or repair the installation using its original method or package manager; pip users can run
python -m pip checkto check dependency requirements (fixes Issue #525).Web API: each object added through the Web API, e.g. from DataLab-Kernel or a Jupyter notebook, showed “Remote request failed: Object ‘…’ not found” in the status bar, although the object was added. Checking that an object name is free is no longer reported as a failed request (fixes Issue #528).
Web API: when replacing an existing workspace object, a failed addition of the replacement could delete the original object. The original object now remains available if the replacement cannot be added (fixes Issue #561).
Deleting objects: deleting many objects at once, or clearing the workspace, froze DataLab for a long time when many curves were displayed (e.g. about 20 seconds for 600 signals). It now takes about one second (fixes Issue #528).
Plugins: after running a processing feature of a plugin whose computation functions are not in the module of its plugin class, such as the plugins created with
datalab-plugin create --project, saving the workspace to an HDF5 file failed with “cannot serialize None value in dict”. These results are now saved, with the name and version of their plugin (fixes Issue #543).Applications panel: when application plugins were installed, docking the floating Applications panel back into the main window could crash DataLab. The panel can now be docked and undocked safely (fixes Issue #527).
Plugin testing: scripts and tests using
datalab_test_app_context, e.g. to test a plugin, sometimes crashed when the Python interpreter exited, after their checks had passed. Their main windows are now destroyed before the Qt application, when the interpreter exits (fixes Issue #556).Development tooling: the VS Code tasks of the source repository reported success even when their command failed (e.g. a failed test, a linting error or a documentation build error), and the tasks chained after them still ran, e.g. updating the translation catalogs after a failed documentation build. These tasks now fail with their command; Coverage full still produces the coverage report when tests fail (fixes Issue #552).
📖 Documentation since version 1.4.0#
Corrected installation and remote-control instructions, descriptions of History and plugins, and tutorial screenshots so they reflect the current application (fixes Issue #569).
Corrected the description of the pulse amplitude, which never falls back on the peak-to-peak amplitude, and documented the bias of the default FWHM method on smooth peaks, the unit of the phase of periodic signals (degrees in the dialogs), the effect of hot pixels on the enclosing circle, of near-zero values on the flat-field correction, how the Web API replaces an object, and the cases where DataLab and DataLab-Web results can be compared. The laser beam tutorial now gives the centroid row used by its screenshots (fixes Issue #571).
Updated the Fabry-Pérot, spectrum and blob-detection tutorial screenshots, the startup page, settings and History panel to match the current interface. Maintainers can regenerate these figures with a single Qt-based script, without manual desktop captures or changes to personal preferences (fixes Issue #520).
Translated card titles in the French documentation now show the same icons as their English counterparts, on the home page and the main section pages (fixes Issue #536).
The DataLab plugin catalog is now presented on the home page, in the navigation bar and in the pages describing the platform, its key features and how to contribute
The plugin documentation is reorganized into an overview page and dedicated pages for installing and managing plugins, deploying them in an organization, developing them, application plugins and the public API; the development page starts with a quick start, “Your first plugin in five minutes”, and presents the plugins of the catalog as examples
New tutorial, “From an idea to the plugin catalog”: create a plugin in a single Python file, try it, then publish it in the DataLab plugin catalog, without any PyPI account
New use case pages present the characterization of scientific cameras, the analysis of transient and pulsed signals, and instrument control and data acquisition, with the Camera & Detector Characterization, Pulse & Transient Characterization and TANGO Instruments plugins of the DataLab plugin catalog; the non-destructive testing page now presents the X-ray Radiography & NDT plugin, also cited as an example in the plugin documentation (Issue #542)
The roadmap is updated: the plugin template is delivered, data acquisition has started, and the milestones of versions 1.1 to 1.4 are listed
The documentation is corrected where it no longer matched the application (Issue #519): menu paths and labels of the tutorials and feature pages, settings and their default values, formulas and results of several processing and analysis features, supported file formats, annotations (stored apart from metadata), the Web API, whose token is not required for local requests by default, the remote control, limited to the local machine, plugins targeting DataLab-Web, macros, installation commands (the conda-forge package is
datalab;[qt]and[opencv]extras), and several examples that no longer workedOn the home page and the Getting started page, the logos of the funders and the icons of the three operating modes are shown again at their intended size, without a white background in the dark theme, and the logos of CEA, CODRA and NLnet link again to their websites instead of opening the image (fixes Issue #521)
The Web API documentation and the tooltip of the Web API localhost bypass setting now warn that, while the Web API server is running with this setting enabled (default), any web page open in the browser, and not only local applications, can access the workspace without a token (Issue #522)
🔧 Improvements since version 1.4.0#
Plugin generator: the release workflow of a generated plugin project (
.github/workflows/release.yml) now checks that the tag matches the version of bothpyproject.tomland the package, runs the tests before building the wheel, and publishes the wheel with its SHA-256 checksum and its build provenance attestation, which users can check withgh attestation verify. A tag such asv0.2.0-rc1publishes a release candidate as a GitHub pre-release, without plugin catalog entry. The generated test no longer compares the plugin version with a fixed value, so it no longer fails when the version changes. Existing plugin projects are not affected.Plugin catalog: a plugin may now require packages that users install themselves in DataLab’s Python environment, such as PyTango for the TANGO Instruments plugin. The Catalog tab shows these packages and offers to install the plugin only once they are installed; in the standalone application, such plugins are shown as not available.
Welcome page: the Applications section is now shown even when no application is installed: it then invites you to discover plugins, and “Discover plugins…” opens the Catalog tab of the plugin configuration dialog, to find and install the plugins of the DataLab plugin catalog; once applications are installed, a “Discover plugins…” button of the section keeps this access. The online documentation and the release notes are now links at the bottom of the page, so that the Walkthroughs column only holds the AI assistant, the guided tour and the demo.
Welcome page: application tiles no longer grow with their description, which made the page crowded once several application plugins were installed (each with a tile for its examples). Tiles now all have the same height and share the width of the page, from the left of Quick actions to the right of Walkthroughs, with as many tiles per row as the width allows: the title stands next to the icon, the description follows it directly, and both are limited (two lines for the title, three for the description), ending with “…” when they are longer, the full text being shown in a tooltip (closes Issue #545).
Plugin instruments: the window of a plugin instrument, e.g. the camera, oscilloscope or radiography simulators, has a new Restore defaults button, which sets the settings of the instrument back to their default values and refreshes the view (closes Issue #545).
Plugin tools: a plugin may declare a tool with
menu=Falseto place it in one of its own submenus: DataLab then lists it in the Applications panel only, instead of also adding it at the end of the plugin’s submenu (closes Issue #545).
DataLab Version 1.4.0#
Requirements#
DataLab-Platform:
Sigima 1.4.0 or later is required: it provides operation contracts and replay qualification, RAW image import, format-aware image export, portable annotations, experimental-source metadata, NPZ object serialization and pixel-center ROI rules.
SigimaX 1.1.0 or later is required: it provides portable PlotPy annotation adapters, high-contrast curve styles, high-DPI splash screen and watermark rendering, and NeXus / multidimensional HDF5 browsing.
DataLab-Capsule 0.2.x (
>= 0.2.0, < 0.3) is required for workspace provenance recording, verification and.dlcapsuleexchange.
PlotPyStack:
guidata 3.17.2 or later is required: it provides histogram-backed range items, module-qualified icon precedence, high-DPI pixmap helpers, and gives every window a title bar following the color theme on Windows, including the windows of plugins (closes Issue #492).
PlotPy 2.12.1 or later is required: it provides sharp SVG icons on high-DPI displays, min/max signal auto-downsampling and faster image overlay interactions.
Third-party:
The Python package and newly generated plugin projects now require Python 3.10 or later. Python 3.9 users must keep an earlier compatible release. The standalone application continues to include its own Python interpreter.
The Windows test runner excludes Python 3.9 from current matrices, while retaining explicit tests of pinned historical releases.
pip 22.2 or later is required to resolve and install additional plugin wheel dependencies.
NumPy up to 2.5 (
>= 1.22, < 2.6) and SciPy up to 1.18 (>= 1.10.1, < 1.19) are now supported.
✨ New Features#
Welcome page:
A new Welcome page, similar to the one of DataLab-Web, is shown at startup and whenever the current signal or image panel is empty. It gathers the main actions to get started: create a signal or an image, open files, browse an HDF5 file, open an HDF5 workspace, import text data, ask the AI Assistant, take the guided tour, run the demo, read the documentation or browse the release notes
The page is a dockable panel, tabbed with the signal and image views: it remains available after data is created or opened, and can be reopened at any time from the ? menu (Welcome page) or from the View menu
A Show welcome page when the current panel is empty option, available on the page itself and in the General settings, controls whether the page is shown automatically, including at startup
Brightness and contrast:
A new Processing > Exposure > Brightness and contrast operation provides a histogram with synchronized minimum, maximum, brightness and contrast controls, deterministic Auto and Reset actions, ROI-aware remapping, and live preview. One input window is initialized from the first selected image and applied to the full selection, with each result preserving its source image and data type.
Intensity windows may extend beyond the observed data range. Exact floating-point bounds survive editing and reapplication, while declared parameter constraints remain enforced and Cancel preserves the original parameters.
RAW image import:
Headerless RAW image files (
*.raw) can now be opened from File > Open…, File > Open from directory…, drag-and-drop, the AI Assistant (load_filetool) and remote control (load_from_filesandload_from_directorywith an optionalRawImageImportParamargument) (implements Issue #237)The RAW image import dialog defines the pixel data type, image width and height, byte offset to the first image, number of images, byte gap between consecutive images, byte order, and the White is zero option (which displays the imported image with an inverted gray colormap without modifying pixel values). Multi-image RAW files create one image per frame
When several RAW files are imported at once, DataLab asks whether to apply the same import parameters to all of them or to set the parameters of each file (each dialog starting from the values confirmed for the previous file); recorded History actions store the import parameters so that replaying them does not reopen the dialog
Format-aware image export:
Saving an image with File > Save… now opens an Export image dialog adapted to the selected file format, with a live preview showing the image in gray levels as it will be read back from the exported file (including quantization and lossy compression for BMP, PNG, JPEG, JPEG 2000 and TIFF files) (implements Issue #256)
Data preparation covers normalization (none, min-max, percentile or manual bounds), out-of-range behavior (clip or rescale), non-finite value policy (error, clip or replace), gamma correction, inversion and conversion to a data type supported by the destination format; when exporting a floating-point or unsupported image to an integer-only format (such as PNG, JPEG or BMP), min-max rescaling is selected by default, and complex images are rejected when the format cannot store complex values
Format-specific options are exposed and validated for each format (PNG compression level, JPEG quality/subsampling/progressive/optimization/smoothing, JPEG 2000 quality mode/layers/irreversible/progression/resolutions/tiling/PLT markers, TIFF compression/level/predictor/rows per strip/tiling/resolution/photometric interpretation, text and CSV delimiter and precision, MAT-File compression), and recorded History actions preserve the export parameters and target object UUIDs
Live processing preview:
Parameter dialogs for compatible 1-to-1 processing now offer an optional live preview for signals and images, including image-to-signal operations. Preview is off by default; Cancel leaves the workspace unchanged, while OK reuses a completed current preview for a single source and otherwise performs the normal processing on the original selection.
Completed previews keep their dedicated process ready for later dialogs, avoiding repeated process startup. Cancelling a preview that is still running stops its process so speculative work cannot continue in the background.
Bounded numeric parameters now offer sliders alongside precise text entry. The Processing tab retains Apply and automatic re-processing, with updates deferred until slider release and valid input.
Plugin status:
Click the Plugins: label in the status bar to open Configure plugins…, as with the Web API and internal console status labels. The shortcut is available whenever plugins are globally enabled, including when no plugins are loaded or a plugin fails to load.
Application plugins:
Added a top-level Applications entry that shows the active scientific application plugins in a new Applications panel
Each application displays its description, version, methods, tools, and examples; processing-only plugins remain outside this focused view
Each method shows the inputs it expects (object type and count, required metadata) and a live status telling whether the current selection can be analyzed, and why not
When required metadata are missing, the status points to Edit > Metadata > Add metadata…, whose Known keys list also offers the keys expected by application methods; the status is updated as soon as objects are modified
Run… starts a method on the current selection directly when the selection is ready, and asks for the input assignment only when it is ambiguous or invalid; the assignment dialog checks the chosen objects before the run can continue
Examples are listed under the methods they are designed for: Try with this example opens the example, prefills the method parameters, and runs the method once they are accepted. One example may serve several methods, and examples designed for no method are listed as datasets
Application plugins may also list tools, such as wizards or editors, in the Applications panel
Plugin tools also appear in the plugin’s submenu of the Plugins menu, in the Signal or Image panel they work on. A tool that needs a selection is disabled until the right objects are selected, and its button in the Applications panel tells what to select
Application plugins may provide instruments, such as simulated cameras or oscilloscopes, without writing any user interface: DataLab shows a window with a live view and the instrument settings, refreshed as settings change or continuously in Live mode, and Acquire adds the acquired objects to a new group, ready for the application’s methods
The New group option of the instrument window may be turned off to add the next acquisitions to the current group of the panel instead of creating a group for each of them. The option and the current group are taken when Acquire is clicked, so that selecting another group during an acquisition does not change its destination. Each instrument sets the default state of the option, and the choice of the user is remembered for each instrument
Instruments are called outside the user interface thread, one call at a time: a slow instrument, such as a real device waiting for its data, no longer freezes DataLab, and repeated refresh requests made meanwhile are merged. Instruments may also declare
openandclosemethods, called when their window opens and closes (including when plugins are reloaded or DataLab quits), for example to connect to a device and to release it. Calls stay one at a time when a window is closed and opened again while a call is running. Whenopenfails, its message stays in the window and the instrument is opened again before the next refresh or acquisition, for example after a settings changePlugin developers describe recipe inputs declaratively (titles, minimum counts, metadata requirements) and may add binding suggestions and fast input checks; every recipe then runs through DataLab’s generic launcher, without plugin-specific input dialogs
The Applications panel is a regular panel that does not hide the DataLab workspace: the first time, it floats beside the main window, as high as this window, so that the data and their views stay in sight while using it. It may then be docked (for example next to the list of signals or images whose selection it analyzes), tabified with another panel or left floating, and DataLab shows it again where the user left it, with the same size; when it is requested while already shown, without switching to another application, it flashes briefly to show where it is. It remains open after starting a method or opening an example
The panel stays compact and readable when docked: each method takes one row, with a colored dot and a short text telling whether the current selection can be analyzed, and a Run… button; below the list, a Details area gives the full details of the selected method (description, reasons of its status, expected inputs with the description and the key of their metadata, examples). The result of the last run of a method is shown below its row. The sections of the list may be collapsed, and the height of the details set with a splitter: DataLab remembers these choices
When several application plugins are installed, their number is shown at the top of the panel, above the list selecting the application; the documentation and welcome page options of the application are in the menu next to it
Completed application methods select their last generated object, consistently with standard DataLab processing
Application plugins add tiles to a new Applications section of the welcome page: by default, one tile per plugin shows its page in the Applications panel, and plugins may declare additional tiles, for example to open an example directly. Additional tiles are shown when there is room for them, and otherwise move to the menu of the main tile
The welcome page stays tidy when many applications are installed: tiles are limited to a configurable number of rows (two by default), with a last tile showing the Applications panel for the others. Applications may be pinned to the top or hidden, and recently used ones come first; the Applications panel offers the same welcome page options
Plugins may declare an icon, shown in the Applications panel, on their welcome page tile and in the Configure plugins… dialog
DataLab checks the declarations of application plugins (methods, examples, tools, welcome page tiles) when loading them: a plugin with an invalid declaration, for example a method whose plugin version was not updated with the plugin, or an example referring to a renamed method, is not loaded and is listed with its error in Configure plugins…, while DataLab and the other plugins start normally (fixes Issue #484)
Plugin installation:
Online and offline Desktop catalogs now accept single-file
datalab_*.pyplugins alongside wheels. Historical files do not need a packaging project or source changes to be listed; catalog inspection never imports or executes them (implements Issue #419)Known script versions can be updated; unversioned files are identified by SHA-256 and offer explicit replacement when their content changes. Scripts previously installed from a file are recognized, and existing development copies are not overwritten
The local catalog commands handle mixed folders of scripts and wheels. Single-file plugins use libraries already available to DataLab and do not trigger dependency downloads. The documentation now explicitly distinguishes source organization, distribution formats and functional capabilities, including in the French offline guide
Plugins can now be installed from a file, without any Python tool, including in the standalone version of DataLab: the new From file tab of Plugins > Configure plugins… accepts a pure-Python wheel (
.whl) or a singledatalab_<name>.pymoduleBefore installing, DataLab checks the file without running it and shows its name, version, plugin classes, dependencies and SHA-256 digest
Plugins may now include additional dependencies as separate wheels, including compatible native libraries, without changing packages already supplied by DataLab. A complete local catalog can be installed without Internet access; missing files and version conflicts are reported before any partial installation is activated
The confirmation dialog lists the additional packages and their fingerprints. They are shared by plugins of the same Python runtime, activated after restarting DataLab, and kept when a plugin is uninstalled
Installed plugins are listed in the same tab, where they can be uninstalled; a newly installed plugin is enabled after a plugin reload when no additional packages are needed, or after restarting when dependencies were added. A new version of an already loaded plugin is used at the next start
The new Catalog tab lists the plugins of the DataLab plugin catalog, official or from the community, with a search field: Install and Update download the wheel, check it against the catalog and install it after confirmation. The catalog address may be changed to use a private copy
Organizations can distribute their own plugins, or the versions of public plugins they have approved, from local plugin catalogs: folders, typically shared on the network, declared in the Advanced tab or with the
DATALAB_PLUGINS_CATALOGenvironment variable. Their plugins are listed in the Catalog tab before those of the DataLab plugin catalog, and hide them when both offer the same pluginA local catalog folder may hold plugin wheels and single-file modules, listed as they are copied to it, or a
catalog.jsonindex giving display names, deprecations and withdrawn releases;datalab-plugin catalog buildwrites this index while keeping curated metadata, anddatalab-plugin catalog checkreports missing, changed, refused or unlisted artifactsAuthors can prepare an offline catalog with
datalab-plugin catalog prepare: it downloads the additional dependency wheels needed by the reference host without installing them, then writes the catalog. The complete folder can be distributed on a network share or copied to computers without Internet accessA catalog that cannot be read, such as a network share that is not connected or the online catalog on a computer without Internet access, is reported in the Catalog tab without hiding the plugins of the other catalogs
The Configure plugins… dialog opens larger and keeps the size it had when last closed. Its tabs follow the order of use and have explicit titles, described by their tooltips: Installed plugins, Catalog, From file and Advanced
Plugin project generator:
datalab-plugin createnow creates a single-file plugin by default, the simplest way to extend DataLab: it only asks for the plugin name and description, then writes adatalab_<name>.pyfile to the user plugin directory. Once the plugins are reloaded, the plugin adds its submenu to the Plugins menu with a sample processing of the selected signals (or images, with--object-kind image), whose short NumPy function is the only part to edit. The file never overwrites an existing one, and can be shared and installed with Plugins > Configure plugins… > From file. The layered installable project, described below, is created with--project(implements Issue #413)Hardened
datalab-plugin createfrom the Camera pilot feedback: generated projects now separate host-independentcorecode, headlessworkfloworchestration, and Desktop/Webadaptersfrom the first commitGenerated package roots expose stable identity without importing Qt; the installed entry point targets the Desktop adapter while Web support is explicitly marked unsupported
Every generated project includes an executable architecture regression test, contribution and architecture documentation, and a changelog
Long plugin names and descriptions are now formatted so newly generated projects pass their bundled Ruff checks without manual source edits
Generated projects are ready to publish: a release workflow attaches the wheel to a GitHub release when a version tag is pushed and prints the entry to submit to the DataLab plugin catalog;
pyproject.tomlrequires the current DataLab major version and declares the project URLsdatalab-plugin create --projectasks for the GitHub account hosting the project and derives the default plugin ID from it (io.github.<account>.<name>),org.datalab.IDs being reserved for DataLab-Platform plugins
Add metadata:
Edit > Metadata > Add metadata… can now extract the value from the formatted text with a regular expression, for example an exposure time or a shot number read from object titles. Objects without a match are left unchanged unless you ask for an error, and a scale factor converts numeric values (e.g. milliseconds to seconds)
Metadata keys may now contain dots and hyphens, as used by plugin keys, and a Known keys list copies a key already present on the selected objects
Workspace provenance:
DataLab now records how each result was obtained: the processing applied, its parameters, the objects it used and their exact content, and the software versions. This record is saved with the HDF5 workspace and read back when the workspace is reopened, by DataLab or by DataLab-Web, so the history of a result survives saving, closing and sharing
Signal and image processing, operations combining several objects and analyses (geometry and table results) are recorded, each with the identity of its Sigima operation; regions of interest, uncertainties, the X alignment applied to signals on different grids and the global options read by the computation (such as kernel normalization) are part of the record, and signals and images of every numeric type (float32, complex…) are covered. An analysis that finds nothing is recorded as such. A result whose data, calibration or regions of interest were changed after it was computed is reported as altered rather than trusted
Any recorded processing or analysis of a qualified Sigima operation (all of them but the phase spectrum) can be recomputed and every result compared with the stored one: exactly, or within the tolerance declared for the operation when the numerical libraries differ (exact, within tolerance, different, or not verified with the reason). Recomputation is available from the Provenance tab of the Properties panel (Verify computation) and from scripts; it never changes the workspace. When only the regions of interest of an input changed since the computation (for example regions created on the analysed image by a detection), it is recomputed on a copy carrying the recorded regions of interest, so that such an analysis remains verifiable
The History tab of the Properties panel is renamed Provenance (to avoid any confusion with the History panel) and now shows the provenance of the selected object: the recorded execution that produced its current content, its parameters and inputs, and, step by step, how each input was itself produced, as it was at the time (e.g. the sigma actually used, even if the input was filtered again since), with the software that recorded it (DataLab Desktop or DataLab Web). Each step is a card with its parameters, inputs and results, and links next to what they act on: show the recorded step of an input, Go to object, Verify this step, History action and details. Check data compares the object, its direct inputs and the earlier steps shown with their recorded states without computing; deleted inputs are reported at once, modified or unavailable ones after checking. Going to the History action of a step keeps the Provenance tab open
The Processing tab restores the recorded call of the selected result, including operations combining several objects (difference of two signals, average of several images, …): every input is kept in order with its role, parameters can be edited, and processing without parameters offers a Reapply button. When the processing cannot be applied again (e.g. an input was deleted), the tab says why. This also works for workspaces reopened from a file, saved by DataLab or by DataLab-Web
History actions are linked to the provenance of their last successful execution: Show execution provenance in the History panel displays it, and the provenance of a result leads to the History action that produced it, analyses included. In the History panel edit mode, editing one result of a command producing several results changes only its parameters in the action
When a workspace file holding provenance is added to the current workspace (without clearing it), DataLab now says before anything changes that its earlier provenance will not be imported: in the question about clearing the workspace (files are chosen first), or in a single warning when that question is not asked. Added objects never inherit the provenance of another object (they are neither shown nor used as the original input of a recorded step, including when verifying it), and their later processing is recorded
DataLab Desktop and DataLab-Web record the same processing identically and replay each other’s results
Before interpolating signals whose X grids differ, DataLab still asks for confirmation; the alignment it applies is the one it has always applied, now named and recorded so that a recomputation applies it identically. Several selected pairs of a two-signal operation now each align the operand on the grid of their own source signal
Workspaces saved by earlier versions keep their processing history as stored: DataLab shows what is known and what is missing, invents nothing, and never recomputes such entries
A workspace whose provenance record is damaged is refused with an explicit message, and the current workspace is left unchanged
Workspaces can be exchanged as capsules (a workspace plus a description of its provenance) from File > Workspace capsule…, the same dialog as in DataLab-Web, so that someone else can check and replay them with either edition:
Export the workspace, with a name, a description, light attachments (protocol, notes, small tables) and, for a workspace opened from a capsule (also after it was saved and opened again), the capsule it is a revision of; the workspace can also be referenced by its publication address instead of being included
Inspect a capsule without opening it: its activities, its attachments (saved only on request, never opened), its integrity (workspace and stored states checked, nothing recomputed) and the verification reports and reviews supplied with it, each checked against the capsule; write a review of it, or Open as workspace (recorded in the History panel)
Local replay recomputes a recorded activity of the open workspace and compares it with its stored result, without changing the workspace; the verification report can be saved to be shared with the capsule
Revisions lists the capsules exported, inspected or retrieved on this computer and compares any two of them (scientific, presentation, storage, environment and attachment changes)
Registry connects to an optional capsule registry to publish capsules (never their workspaces) and reviews, list the published revisions and retrieve them, everything returned being checked
Scripts and macros can export and open capsules too (
export_capsuleandopen_capsuleof the proxy), and theworkspace-capsulecommand line of DataLab-Capsule still validates, compares and verifies themA capsule, or the workspace it references, that exceeds the size limit is refused before being read, and DataLab never writes a capsule it would refuse to open, nor writes the capsule or its referenced workspace over each other or over an attachment; the limits and the measured costs are given in the Workspace capsules page
The standalone version (installer) embeds what capsules need (schemas and RO-Crate context of DataLab-Capsule)
Object references in titles:
DataLab now uses stable UUID-based object references across object trees, result titles, plot legends and HDF5 workspaces (implements Issue #367 and Issue #149).
Signal, image and group items in the trees now show their current title followed by their own stable
#UUID8reference (#gUUID8for groups), replacing the formers001,i001andgs001identifiersSource references embedded in result titles no longer change when objects are reordered, removed or renumbered, and remain clickable to select the source object
The new References in result titles setting lets you display these references either as 8-character UUIDs (default) or as the current source titles, which follow later renames and are resolved through whole processing chains (e.g.
derivative(derivative(My signal)))When a source object is deleted, its reference falls back to its 8-character UUID instead of becoming an anonymous placeholder
Titles of workspaces saved by previous versions keep their former
s001-style references as plain text
Signal and image lists:
Signals, images and groups can now be deleted directly from the lists, as in DataLab-Web (implements Issue #449).
A delete button appears at the right of the hovered signal, image or group: it turns red when the mouse cursor is over it, and its tooltip tells what it deletes
Clicking it deletes the item, or the whole selection when the item is part of it, after confirmation. Deleting an item which is not selected keeps the current selection and the displayed objects unchanged
As with the Remove action, the deletion is recorded in the History panel
The button may be hidden with the new Show delete button when hovering list items setting (Visualization > Common)
Curve fitting formulas:
Interactive fitting dialogs now show the formula of the fitted model below the fit parameters, and each parameter label shows its symbol in the formula (e.g. “Scale factor (x₀)”): the parameters stored in the fitted signal metadata can thus be related to the model without reading the source code. The Fitting documentation also gives, for each model, the name under which each parameter is stored (fixes Issue #400)
Polynomial and linear fit coefficients are labelled
a,b,c… as in the formula and in the metadata, instead ofc2,c1,c0The Fitting documentation explains how to derive the peak wavelength, the temperature and the emissivity from the parameters of a Planckian fit
Remote control and plugins:
The Contour isoline plot plugin (
org.datalab.isolevel-map, version 1.1.0), formerly provided as a plugin example, is now a built-in plugin available directly from Plugins > Contour isoline plot > Show contour plot…add_groupnow returns the UUID of the created group, so that a remote client, a macro or a plugin can add objects to it withadd_object(obj, group_id=...)(oradd_signal/add_image) instead of looking the group up by its titleload_from_filesandload_from_directorynow accept an optionalparamargument with format-specific import parameters (e.g.RawImageImportParamfor RAW images), applied without showing any dialog
NeXus files:
NeXus
NXdatagroups are shown as single items in the HDF5 browser and imported as signals or images with their axes, units, labels and, for signals, uncertainties. Bin edges are converted to bin centers, non-uniform image axes are kept, and decreasing image axes are reversed with the image. Invalid axes are reported in the browser instead of being guessedFiles written with the legacy NeXus conventions, such as the scans of synchrotron SOLEIL, are imported with their axes (
axisandprimaryattributes of the axis datasets), and every counter and detector of a scan is shown as an item of its own: counters as signals versus the scanned motor, detectors as images (a spectrum per point) or image stacks (an image per point) with the motor along their first axisThe HDF5 browser selects and previews the default measurement of a NeXus file (
defaultattributes) when it is openedCompressed detector data, such as the bitshuffle/LZ4 data of Eiger detectors, is now read (with hdf5plugin, also in the standalone application); a dataset compressed with an unavailable filter is reported in the browser instead of failing
Imported signals and images record where they come from (file, group, signal and axes, applied conversions, title and times of the NeXus entry) in their
experimental_sourcemetadata, which is saved with the workspace: a saved workspace reopens without the NeXus file.nxsand.nx5files are proposed when opening HDF5 filesPlugins can describe the signals and images of files that NeXus
NXdatagroups designate incompletely (no axes, several counters…) with declarative data mappings (JSON) registered withsigimax.h5.mapping.register_mapping: the HDF5 browser then shows each described object as an extra item, imported with its axes, labels and units; the axes of a mapped object may describe its first dimensions only, so that one mapping applies to the counters and detectors of a scan
Installation check:
The new
--smoke-testcommand line option checks that DataLab is correctly installed without opening its main window, e.g.DataLab.exe --smoke-test report.jsonfor the standalone version ordatalab --smoke-test report.json: the packages it relies on, the resources provided with it (macro templates, PDF documentation, French translations and icons) and processing in a separate process. The results are written in a JSON report, and the exit code is not zero when a check failed (see Command line features)Before being published, each Windows installer is now installed on a test computer, where the installed application is started and checked this way (implements Issue #366)
🔄 Changes#
Curve readability:
Signal curves are easier to read, especially when many curves are shown (closes Issue #453):
Curve colors now stand out on both light and dark backgrounds: several colors of the former palette (orange, pink, yellow-green, cyan) were pale on a white background, and gray curves could not be told apart from the grid
After the first 10 curves, curves are now drawn with dashed lines only: dash-dot and dash-dot-dot lines, which drew only half of the curve, are no longer used
The plot grid now uses discreet solid lines instead of gray dotted lines, which competed with dashed curves; it also follows color theme changes
Curve colors and line styles now depend on the position of the curves among those shown on the plot, in the order of the signals in the panel: the first curve shown is always drawn in the first color. Opening the ROI editor, View in a new window or other dialogs no longer shifts the styles of the next curves, which could reach dotted styles with only a few curves shown
Automatic curve styles are no longer saved in workspaces: only a color or a line style chosen by the user is saved and always kept, whatever its value. When opening workspaces and history files saved with DataLab 1.3 or earlier, which saved the automatic style of every curve, these styles are removed so that curves are displayed with the new automatic styles; a style chosen with these versions is removed as well when it is identical to one of their automatic styles. Metadata imported from a file are applied as they are
View > Reset curve styles now gives the selected signals automatic curve styles again, removing the colors and line styles chosen for them
On high-DPI screens (e.g. with a 150 % display scaling), curves of width 1 are now drawn one screen pixel wide, which makes their rendering much faster: noisy signals were drawn up to several hundred times slower than at 100 % (requires PlotPy ≥ 2.12.0)
Synthetic images in physical coordinates:
The 2D Gaussian, 2D ramp, sinusoidal grating, ring pattern, Siemens star and 2D sinc images were computed over the [xmin, xmax] × [ymin, ymax] domain of their creation parameters, but were displayed in pixel coordinates: their centers, widths and spatial frequencies did not match the axes (a sinusoidal grating of frequency 0.1 over [0, 100] showed 10 periods along an axis graduated from 0 to 1023). These images are now displayed over their domain. The new Physical coordinates option of their creation parameters restores pixel coordinates when disabled (fixes Issue #400)
Images created with earlier versions keep their pixel coordinates when their creation parameters are edited or recomputed
Guided tour:
The guided tour no longer starts automatically at first launch: it is now offered from the welcome page, and remains available from the ? menu
Large campaign responsiveness:
Opening, selecting, and displaying large generated campaigns no longer repeats a complete workspace and plot refresh for every object
Native HDF5 workspaces and multi-object recipe results now load in groups, making large signal and image collections available substantially faster while preserving every selected and visible object
Image display responsiveness:
Moving or resizing regions of interest, annotations and markers over images is now much smoother, as the image is no longer rendered again for each mouse move: with a 2048 × 2048 image, each mouse move takes about 5 ms instead of 13 ms (13 ms instead of 33 ms with a 200% display scaling). Panning and zooming images without masked pixels are also up to 25% faster (requires PlotPy ≥ 2.12.1)
Sharp icons with display scaling:
The icons of the plots and of the editors provided by PlotPy and guidata (plot toolbars, contrast panel, cross sections, item list, array editor, …) are now SVG images: they remain sharp with a display scaling such as 125 %, 150 % or 200 %, instead of looking blurry (requires PlotPy ≥ 2.12.1, closes Issue #470)
The Processing > Resize action of images showed the column resizing icon of guidata’s editors instead of its own icon: DataLab icons now take precedence over guidata and PlotPy icons with the same name (requires guidata ≥ 3.17.1, closes Issue #472)
Plugin icons (
PluginInfo.icon, tools and welcome tiles) may also designate an icon registered by a module, e.g."datalab:delete.svg", in addition topackage:pathresources and DataLab icon file names
Portable plot annotations:
Plot annotations are now stored in a renderer-independent format shared with DataLab-Web, so annotations in workspaces and
.dlabannfiles are no longer tied to PlotPyExisting PlotPy annotations remain readable and are converted only after an annotation edit is accepted; simply opening a workspace or cancelling the editor leaves its data unchanged
Annotation identifiers, lock state, custom metadata and extension data are preserved across edits, while unknown third-party payloads are retained without modification
Rectangles and segments drawn on signals are now the same labeled shapes as on images; unlike on images, segments drawn on signals do not show their length and angle, which would mix X and Y units
Regions of interest:
A rectangular region of interest now selects the pixels whose centers lie inside the rectangle, bounds included, like circular and polygonal ROIs and like the display: its bounds were rounded to the nearest pixel, which could shift the selection by up to half a pixel. A rectangle whose bounds lie exactly on pixel centers now includes the pixels on both bounds, i.e. one more row and one more column than before. This applies to processing and analysis in ROIs, to ROI extraction and to the background area of the offset correction
Circular and polygonal regions of interest now also select the pixels whose centers lie inside their shape, boundary included: polygon vertices were rounded to the nearest pixel, so that the selected pixels could differ from the polygon drawn on the image, and circles were drawn in pixel units on images with non-uniform coordinates
A rectangular ROI defined in pixel indices now covers its pixels up to their outer edges, so that it keeps the same pixels when converted to physical coordinates and back
The cells of ROI grids (ROI > Create ROI grid…) are now made of whole pixels, so that adjacent cells never share a pixel and a full-size grid covers every pixel exactly once, whatever the image size; on images with non-uniform coordinates, the grid now spans the whole image, whose last rows and columns were left out
Regions of interest saved with earlier versions are not converted: they select pixels according to these rules when they are used again, so that processing and analysis in these regions may give slightly different results when recomputed
Image profiles in physical coordinates:
Line, segment, average and radial profiles (Analysis > Intensity profiles) are now defined in the physical coordinates of the image, like regions of interest and analysis results, instead of pixel indices: the profile parameters show coordinates in the units of the image axes, and the X axis of the resulting signals holds physical coordinates or distances with the matching unit. On images without calibration (origin 0, pixel size 1), results are unchanged
The radial profile is now circular on images with rectangular pixels or non-uniform coordinates, and its center is correct on calibrated images, where the image center was misplaced
Segment and radial profiles are refused on images whose X and Y units differ, since distances are then undefined; coordinates outside the image are refused instead of being silently moved to its edge. Lines and segments drawn beyond the image in the profile dialog are brought back to its outermost pixels
Scripts, macros and plugins must use the new parameter names (
y/xfor line profiles,x1,y1,x2,y2for segment and average profiles,xc/ycfor radial profiles) and physical coordinates. Profile computations recorded in the provenance of workspaces saved with earlier versions are reported as not verifiable, instead of being replayed with a different meaning
HDF5 browser:
Browsing an HDF5 file no longer reads its datasets entirely: they are listed and described from their shape and data type, and the value previews read only their first elements. Large files open faster and use less memory; data is read when a signal or an image is previewed or imported
The HDF5 browser stays responsive while a preview is read, and items selected meanwhile are not read; selecting an item that cannot be previewed (a group, for example) clears the preview. Ordinary 1D arrays and 2D images, NeXus and mapped objects above 2²⁴ values (about 16.8 million) are previewed with a step along each axis (the axes of NeXus images are still read entirely, to check them as on import); compound datasets and signals stored in 2D tables still use complete objects for graphical previews
Datasets with three dimensions or more (image stacks, spectral cubes…) are now listed in the HDF5 browser and imported as one image or signal per frame: a view selector chooses the displayed axes, the frames (first, last and step along another axis) and the index of the other axes, previews the first frame, and imports the frames once Import these frames is checked (with a confirmation beyond 100 frames). Frames of NeXus datasets keep their axes, and their title gives their position (e.g. the motor position of each detector image of a scan)
Dataset attributes with more than 10 000 values are no longer copied to the metadata of imported objects
See the HDF5 browser guide for scientific object import, experimental origin and preview limits.
Scripts, macros and plugins:
An object or a group looked up by its title, from a macro, a plugin, a remote client (e.g.
get_object("My signal")) or the Web API, is no longer silently replaced by the first one found when several objects or groups share this title (e.g. files imported twice, “Untitled” objects): an error now names the matching objects with their UUID, which identifies them unambiguously (HTTP 409 for the Web API)For plugin developers:
patch_title_with_idsno longer takes the third argument (id_func), and the helpers of the formers001-style identifiers (SHORT_ID_REGEX,find_by_short_id,find_short_ids_in_title,replace_short_ids_by_uuids_in_titles,replace_uuids_by_short_ids_in_titles) are removed fromdatalab.objectmodel, as titles now refer to objects by their UUIDFor developers of custom DataLab controllers derived from
BaseProxy: the newexport_capsuleandopen_capsulemethods must be implemented
🛠️ Bug Fixes#
Performance with many objects:
Fixed processing becoming progressively slower as signals or images accumulate in a panel: each new result rewrote the titles of every existing object, rebuilt the whole object tree and measured every title again, so creating 1000 signals one after the other took about 39 minutes; it now takes about 3 minutes (a processing step with 1000 objects in the panel dropped from 2.3 s to about 0.3 s)
Image panels no longer slow down because of cross-section curves kept for hidden images (requires PlotPy ≥ 2.11.2)
Fixed titles referring to objects numbered 1000 or above (e.g.,
s1000) being corrupted when objects were added or reordered
Window layout:
Fixed the width of the signal/image panels not being restored at startup: after narrowing the panels and restarting DataLab, they came back wider (up to their maximum width) instead of keeping the width chosen by the user
Fixed DataLab failing to start when its configuration file held an invalid window layout or an invalid log file name (e.g., after editing the file by hand): such settings are now ignored, and the default ones are used
Console:
Fixed the console still being shown on errors after disabling Show console on error in the Settings dialog (with Qt 5, when the setting was enabled at startup), until DataLab was restarted
Demo:
Fixed the demo failing with the error “No module named ‘pytest’” in the standalone version of DataLab (installed with the MSI installer) and in installations without the test dependencies (e.g.,
pip install datalab-platform), whether started from the ? menu, the welcome page, the last step of the guided tour or thedatalab-democommand (fixes Issue #365)The
datalab-democommand no longer deletes the DataLab configuration of the user: the demo now runs with the default settings, without reading or changing the settings of the user
Standalone version:
Fixed the standalone version of DataLab exiting with a success code when it failed to start: it now exits with an error code, so that scripts and deployment tools launching it can detect the failure (the cause is still written in
datalab_error.log, in the current folder)
Display scaling:
Fixed the splash screen, the DataLab logo of the welcome page and of the guided tour, and the watermark of empty plots being blurry with a Windows display scaling greater than 100% (e.g., 125%): they are now displayed at the resolution of the screen
Fixed most of the text of the user interface (welcome page, object properties, panel tabs, dialog boxes, …) keeping its 100% size with a Windows display scaling greater than 100% (e.g., 125%), while menus and the status bar were enlarged. All the text now follows the display scaling and the Windows text size setting, and uses the standard font of Windows applications (“Segoe UI”). This affected DataLab running with Qt 5, like the standalone application (requires guidata ≥ 3.16.1)
Fixed the line of the X axis of signal and image plots not being drawn with a display scaling such as 125% or 150%, depending on the size and position of the plot in the window (requires PythonQwt ≥ 0.17.1)
Fixed images being displayed with a lower resolution than the screen with a display scaling greater than 100%: with a 200% scaling, images were shown with half the resolution of the screen, and with 125%, one screen pixel row and column out of five were duplicated, which made fine details (e.g., noise) look irregular. Images are now displayed at the resolution of the screen (requires PlotPy ≥ 2.12.1)
Fixed the red plugin icon of the status bar (shown when some plugins failed to load) being too small and shifted to the top-left corner with a display scaling greater than 100% (e.g., half its size with 200%)
Dark theme:
Fixed most windows keeping a white title bar with the dark theme on Windows, next to their dark content: the conversation history of the AI Assistant, the workspace capsule, the command palette and other dialog boxes, the import wizard, the confirmation and input boxes (e.g., when renaming a group or a conversation) and the panels detached from the main window (signal and image views, macro, history, AI Assistant and welcome panels, console). All the windows now have a dark title bar with the dark theme, also after changing the color mode in the Settings dialog (requires guidata ≥ 3.17.2, fixes Issue #490)
Plot watermark:
Fixed the watermark of empty plots being read from another installation of DataLab sharing the same user configuration (e.g., DataLab installed in another Python environment): the watermark disappeared once this other installation was removed, and stayed blurry on high-DPI screens when it was an older version. The watermark of the running DataLab is now always used
Signal display:
Fixed auto downsampling of signals (Settings > Visualization > Signals) only starting above 5 times the configured maximum number of points: with the default value of 100,000 points, signals of a few hundred thousand points were drawn in full, which slowed down the display. Signals with more points than the maximum are now downsampled so that at most this number of points is drawn, and the Settings dialog and the documentation now give the actual default value (100,000 instead of 10,000) (fixes Issue #455)
Fixed changes of the auto downsampling and line width settings only applying to the curves of the selected signals: the other curves kept the former settings, even once selected again. These settings now apply to all curves as soon as the Settings dialog is closed
Fixed the X axis ticks being shifted relative to the vertical grid lines and the curves after opening a signal or changing the selection, for signals whose X axis does not start and end on a graduation (e.g.,
step.npyfrom the Sigima test data): the shift remained until the view was panned or zoomed. The X axis now stays aligned with the grid and the curves (fixes Issue #457)Fixed curves of reloaded workspaces keeping the line width they had when the workspace was saved: changes of the line width setting (Settings > Visualization > Signals) no longer applied to them, and thick lines were no longer drawn 1 pixel wide for signals with many points, which slowed down their display. Only line widths chosen for a curve (e.g. in its parameters dialog) are now saved, and they are also drawn 1 pixel wide for signals with many points. When opening workspaces and history files saved with DataLab 1.3 or earlier, which saved the line width of every curve, line widths of 1 are removed so that curves follow the line width setting; other line widths are kept
Custom colormaps:
Fixed custom colormaps (created with the colormap manager) being saved outside DataLab’s configuration directory, in a
.plotpyor.nonedirectory of the user’s configuration directory shared with other applications based on PlotPy, depending on how DataLab was started. They are now saved in theplotpysubdirectory of DataLab’s configuration directory; the first time, custom colormaps saved in the former directories are copied there (requires PlotPy ≥ 2.12.0)
HDF5 import:
Fixed images with unsigned 32-bit values above 2 147 483 647 being clipped when imported from an HDF5 file (“Clipping uint32 data to int32” warning): such images are now imported as 64-bit floating-point images keeping every value exactly. Other integer images that no supported integer type can hold also use a sufficiently precise floating-point type when possible, instead of rounding to 32-bit floating-point values beyond 16 777 216; a warning is emitted when no supported type guarantees exact conversion (in particular beyond ±2⁵³ for float64)
Fixed single-precision complex (
complex64) 1D datasets losing their imaginary part when imported from an HDF5 file: they are now imported as complex signals. Integer signals beyond ±2⁵³ now also emit the precision warning instead of being rounded silentlyFixed the import of several datasets from HDF5 files stopping at the first dataset that could not be imported: the other datasets are now imported, and the datasets that could not be are listed in a single message at the end
Fixed HDF5 files containing broken links failing to open in the HDF5 browser, for example synchrotron master files whose detector files are not available (external links to a missing file): the broken links are now skipped and the rest of the file can be browsed and imported
Fixed HDF5 files containing a group linked into itself or into one of its subgroups failing to open in the HDF5 browser (the file was browsed endlessly until an error): such links back to a parent group are now skipped, while other aliases of a group or dataset are still shown at each of their paths
Fixed datasets stored in other files, such as the detector data of synchrotron master files, missing from the HDF5 browser even when these files are available: the datasets reached through an external link were looked up at their path in their own file, so that they were skipped, or the data of another dataset of the opened file, found at that path, was imported instead. They are now listed and imported at their path in the opened file
Fixed HDF5 files whose name contains a comma failing to open: the comma was taken for the separator between a file name and the name of a dataset to import. When appending such a workspace, the question about clearing the workspace also did not mention that its provenance is not imported, and an append cancelled by the user was still recorded in the History panel
Fixed datasets holding a single row of values (e.g., a single scan stored as a 1 × N array) being silently skipped when imported from an HDF5 file: they are now imported and previewed as signals
Fixed an error when clicking Show array in the HDF5 browser after selecting a group with the keyboard, or after closing the file of the selected dataset: the button now only shows arrays, and is disabled once the file is closed
Workspace saving:
Fixed a failed save destroying the previous version of the workspace file: the file was emptied before the data was written, so that an error during the save, a full disk or an interruption left neither the previous nor the new workspace on disk. Workspaces, capsules and History files (
.dlhist) are now written to a temporary file, which replaces the previous file only once completely written; if the save fails, the previous file is left untouched
File import:
Fixed compressed TIFF files (LZW, JPEG or WebP compression) failing to open with the error “requires the ‘imagecodecs’ package”; LZW is the default compression of many acquisition and image-editing tools. These files now open out of the box, including with the standalone installer. CCITT-compressed bilevel images are also supported with the standalone installer and with Python 3.11 or later (fixes Issue #381)
Metadata:
Adding, pasting or deleting metadata now marks the workspace as modified, so DataLab asks to save these changes before closing
Importing metadata from a file into an object (Edit > Metadata > Import metadata) no longer gives it the identity of the object the metadata were exported from: two objects sharing an identity confused source references in titles, Select source objects and the saving of the workspace
Regions of interest:
Fixed the Inverse ROI logic option of image ROIs being lost when validating the ROI editor: the option had no effect when set in the coordinates dialog of the editor (e.g., Rectangular ROI), and image ROIs created with it (e.g., by a macro) lost it once edited
Fixed ROIs defined in pixel indices (e.g., by a macro or with the remote control) being converted to physical coordinates when the ROI editor was validated. They now stay in pixel indices, also when moved in the editor
Processing:
Results of the extraction of several regions of interest into a single object can be processed again: the Processing tab showed “Unknown computing feature: extract_rois” with Reapply disabled, and replaying the History skipped the action
Edit > Recompute now also recomputes results of operations combining several objects (e.g., difference of two signals, average of several images), which were skipped
Fixed Processing > Resampling of images shifting every value by half a pixel: resampling an image with the default parameters, i.e. onto its own grid, filled its last row and column with NaN. The X/Y min/max parameters are now the coordinates of the centers of the first and last output pixels (by default, those of the image), and the ‘Pixel size’ mode keeps the requested pixel size exactly
Fixed the sum, average, product and standard deviation of several signals, and their combination into an image, depending on the order of the signals: when a real signal came first, the imaginary parts of the complex signals were discarded
The sum, average and product of several images now accept complex images, which raised an error
Fixed Processing > Resampling of complex images with the default parameters, and the Z-axis Polynomial calibration of complex images, discarding the imaginary parts: complex images now stay complex
Fixed modifying the result of a diagonal, horizontal or vertical flip or a 90° rotation of an image in place (e.g. from a macro or a plugin) also modifying its source image
Fixed the extraction of several regions of interest into a single image giving wrong coordinates to images with non-uniform coordinates (the X coordinates were taken from the extracted rows and the Y coordinates from the extracted columns)
The arbitrary rotation of images (Processing > Geometry > Rotate by…) now refuses images with non-uniform coordinates, like resizing and resampling, instead of producing an image whose coordinates did not match its pixels
Fixed the uncertainty of the product of signals being lost where one of the signals is zero: it is now propagated there too
Fixed ROIs, annotations and analysis results (centroids, detected peaks, contours…) being shifted by one pixel after a flip or a 90° rotation of an image; after a 90° rotation of a non-square image, or of an image whose coordinates do not start at zero, analysis results even ended up far from the features they describe. They now move exactly with the pixels; for images with non-uniform coordinates, circles, ellipses and polygons are moved approximately when the pixels are reversed along a non-uniform axis (a circle does not stay a circle). ROIs defined in pixel indices are also moved correctly, and inverse ROIs (which mask the inside of their shape) no longer become normal ROIs after these operations, a translation or a distribution of images on a grid
Fixed ROI > Extract… giving the extracted image a wrong origin when the coordinates of the source image do not start at zero, and coordinates not matching the extracted pixels for images with non-uniform coordinates
Fixed the coordinates of analysis results computed in a ROI being off by up to half a pixel
Fixed the analysis results of images with non-uniform coordinates (Analysis > Centroid…, Minimum enclosing circle center…, 2D peak detection, Contour detection…, Circle Hough transform… and Blob detection) being given in pixel indices, or according to the former pixel size and origin of an image switched to non-uniform coordinates, so that they were drawn away from the features they describe: they are now placed at the positions of the pixels on the image axes, with radii scaled by the local pixel size. The results of images with uniform coordinates are unchanged
ROI > Extract… now reports an explicit error for a region of interest holding no pixel of the image (for example a cell of a ROI grid narrower than a pixel, or a region outside the image) instead of silently creating an empty image; the other regions of interest are still extracted
Fixed Processing > Level adjustment > Offset correction using a background area truncated to integer coordinates: the area was wrong for images whose coordinates are not in pixels (e.g. in millimeters), and empty when smaller than one unit
Plugins:
Plugins organized as a folder, such as those created by
datalab-plugin create --project, are now loaded when their folder is added to the plugin directories, without being installed first: they were silently ignored, unless an earlier installation had left package information in the project. The project folder itself can now be added, instead of itssrcsubfolder (fixes Issue #411)Fixed small empty windows briefly flashing on screen when opening Plugins > Configure plugins…, just before the dialog appeared (fixes Issue #486)
Plugins and History:
Replaying a History action or re-processing a result created by a plugin feature that shares its name with a built-in feature now runs the plugin feature instead of the built-in one
History Panel:
Fixed an oversized warning appearing when deleting a group of signals or images whose processing results were kept in another group, even when the History panel was not recording: it listed every deleted object, often as an unreadable identifier, and stated that “operations could not be reconnected” although there was nothing to reconnect. Deleting such objects now shows no message, and the results depending on them are left unchanged as before. The warning is now shown only when a recorded processing step loses its source, and it lists each step once, in a message of bounded size (fixes Issue #451)
Fixed the Description column of the History panel showing the raw content of voluminous parameters, e.g. the whole histogram of a Brightness and contrast action: the row grew to a huge height and scrolling the panel became erratic. The description now shows the actual parameters (e.g. the minimum and maximum of the brightness and contrast range) on a single line, shortened to the column width with the full summary in its tooltip; long values are shortened, and expanded details scroll inside a row of limited height (fixes Issue #369)
Object titles:
Fixed plot legends still showing the original generated title after renaming an object, e.g. a profile extracted from an image (fixes Issue #323)
Fixed source references in titles being unreadable when the object is selected in the signal or image tree, notably on Linux (fixes Issue #358)
Fixed result titles pointing to the wrong source objects after appending an HDF5 workspace to a non-empty session; appending the same workspace twice now also keeps each imported processing chain independent, so Recompute and Select source objects use the sources of the same import (fixes Issue #357)
Curve fitting:
Fixed the automatic Planckian fit (Processing > Fitting > Planckian fit) returning a curve about 150 times lower than the signal on data in physical units, such as a blackbody signal with wavelengths in meters; the interactive Planckian fit dialog also starts from a curve matching the signal peak (fixes Issue #400)
Fixed exponential fitting failing on short time constants depending on the X unit, e.g. a 1 ms time constant with X in seconds (automatic and interactive fits)
Signal creation:
The blackbody signal (Create > Blackbody (Planck’s law)) now has axis titles and units (wavelength in m, spectral radiance in W·sr⁻¹·m⁻³), which were missing (fixes Issue #400)
Remote control:
Fixed files loaded remotely with
load_from_filesorload_from_directorybeing sometimes silently ignored, or still loading after the call had returned, when requested right after DataLab started: requests received while DataLab is starting are now processed as soon as its main window is ready (fixes Issue #398)Errors raised by remote commands, such as
calcwith an unknown computation function orset_objectwith an unknown object, are now reported to the client instead of being silently ignoredFixed remote commands sometimes returning the result of the previous command instead of their own: for example,
get_current_object_uuidcould return a panel name instead ofNonewhen there was no current objectFixed DataLab freezing when closed while a remote command was still running (e.g. when closing the main window with
call_method("close"))Fixed DataLab crashing, or opening an error message box, when a request of a remote client or a macro failed, e.g. a computation requested with
calcthat failed on the selected signal, or a file that could not be read withload_from_files: the client was not informed of the error, and DataLab crashed when it ran without display (e.g. in the background, for tests or automation). The error is now reported to the client, as anxmlrpc.client.Faultwith XML-RPC or as an HTTP error with the Web API (500 for a failed computation, 404 for an unknown computation function), and shown in the status bar of DataLab; this also applies to the datasets of an HDF5 file that could not be imported. Macros and remote clients now get an exception where they continued after the message box was closed (fixes Issue #404)Fixed every Web API request failing after 30 seconds with an HTTP 500 error (“Main thread execution timed out”) when the Web API server had been started by a remote client with
start_webapi_server(XML-RPC), including requests that should have failed at once, such ascalcwith an unknown computation function; the requested operation could even be carried out later, after the error had been returned. Requests are now served as when the server is started from the File > Web API menu (fixes Issue #403)Fixed images with non-uniform coordinates losing them when transferred through the Web API (
GETandPUT /objects/{name}/data): they were rebuilt with default uniform coordinates. The NPZ data format now includes these coordinates; archives written by earlier versions remain readable. Clients with their own NPZ reader, such as DataLab-Kernel, must support the new entries to take advantage of this
AI Assistant:
Fixed the Save to Macros link missing when a conversation is reopened from History…: a macro proposed by the assistant in a previous session could no longer be saved. Reopened conversations also showed every tool call as successful, including the failed ones; they now show the actual outcome (fixes Issue #447)
Saving a macro proposed by the assistant now offers a Show in Macro panel link: the Macro panel is tabbed with the AI Assistant, so the saved macro stayed hidden behind the conversation
Fixed the assistant answering a request to write a macro with the code in its reply, which could not be saved to the Macro panel: it now proposes the macro without running it (the confirmation dialog states that it will not be executed) and offers the Save to Macros link. A Save to Macros link is also offered under each Python code block of its replies, unless macro creation is disabled
Security:
Opening a workspace received from someone else can no longer make DataLab load or run other code installed on the computer. The parameters stored with objects, analysis results and History actions name their class, and DataLab imported whatever module they named and called the object named there, which a modified file could use, for example, to make DataLab quit. Only parameter classes of DataLab, Sigima, guidata and installed plugins are now used; other stored parameters are reported as unreadable (fixes Issue #407)