Skip to content

computation-engine.md

Latest commit

 

History

History
71 lines (56 loc) · 3.43 KB

computation-engine.md

File metadata and controls

71 lines (56 loc) · 3.43 KB
Raw

Computation engine (Sigima)

The engine that actually computes signals and images is Sigima, a headless Python library shared between desktop DataLab and DataLab-Web. Whatever the host application, the algorithms, parameters and numerical results are the same.

Where the documentation lives

There is no separate Sigima documentation for the Web, by design. The existing reference on the DataLab project website — even though its screenshots show the Qt desktop UI — describes the same parameters and the same outputs as those exposed here.

Useful entry points (desktop-framed but Web-relevant):

How processings appear in DataLab-Web

The menu bar is populated automatically by introspecting Sigima's catalog (build_signal_catalog() / build_image_catalog()). When a new processing is added to Sigima upstream, it shows up in DataLab-Web at the next release without any UI code change.

Parameter dialogs

Each processing's parameters are described by a guidata.DataSet schema. DataLab-Web reads that schema and renders the dialog automatically — the same data model that the desktop application uses to draw its Qt forms.

Eligible bounded numeric parameters also get a slider beside the exact numeric input. The slider never replaces typed entry and is omitted when the schema does not provide safe finite bounds.

Compatible one-input/one-output processings expose an optional Preview checkbox, initially off. Enabling it reuses the already-loaded computation runtime to process a private copy of the source; it does not add an object, alter history, or mark the workspace as modified. Edits are debounced while typing and sampled while dragging a slider, with an immediate refresh when the slider is released. If several objects were selected, Preview source chooses the representative input without changing the selection or the final batch operation.

Cancel and Esc discard the temporary result and any waiting refresh. An already-running computation finishes in the background and its late result is ignored. With one source, OK consumes a completed, up-to-date preview once and publishes it through the normal processing path. A running or stale preview, a changed source, and multi-object or group selections fall back to normal computation from the original source and current parameters. Preview is hidden in the runtime=main fallback to avoid running this interactive work on the UI thread. The Processing tab in the object side panel retains its explicit Apply and Reset controls; editing that tab never republishes a result automatically.

What the desktop docs do not cover

  • The runtime constraints of the browser (memory limits, OPFS, cross-origin isolation) — see Differences from desktop DataLab.
  • DataLab-Web-specific UI affordances (Plotly toolbar, macro / notebook workers, plugin loader).
  • The TypeScript SDK for embedding DataLab-Web in another web app.

For everything else — operators, filters, fits, detection algorithms, units, ROI semantics, HDF5 file format — the desktop documentation is authoritative.