AppBlock example — run a Fiji macro

FijiMacro (block.py) sends an image to Fiji (ImageJ), runs the gaussian_blur.ijm macro on it, and reads the result back. AppBlock turns any file-based external tool into a workflow step — headless (as here) or interactive, where the user edits the data in the app's GUI and the block waits for them to finish.

You declare, the base class runs

The striking thing about AppBlock is how little you write. You override a few class attributes and do not write run():

You set	What it does
app_command	The program to launch (a list of command parts)
input_ports	What data the block accepts
output_ports	What it returns
output_patterns	Glob(s) for the files the app will produce (["*.tif"])

The inherited run() then does all of this for you:

1. Writes inputs into an exchange folder: <exchange>/inputs/.
2. Launches the app, appending the exchange-folder path as the final command argument and validating the command first (no shell injection).
3. Watches <exchange>/outputs/ for files matching output_patterns, waiting until they stop changing (so half-written files are never read).
4. Collects those files back into your output ports.

The data flow, concretely

<exchange>/
  inputs/      <- your "image" input lands here
  outputs/     <- the macro writes here; SciStudio watches for *.tif

The macro learns the exchange path from getArgument() and reads/writes the two sub-folders — see gaussian_blur.ijm. That contract (inputs in, outputs out, exchange path as the last argument) is the same for any app; only the macro/script changes.

Why Artifact ports here

This example uses Artifact (an opaque file) for both ports because Fiji reads and writes TIFF natively, so passing the image file straight through is the simplest, most honest thing. If your input is instead an Array (a SciStudio image), the bridge serialises it to the exchange folder as a NumPy .npy file — fine for an app that reads .npy, but Fiji would need a .npy reader plugin. For Fiji, keep the data as image files (Artifact).

Interactive use

Drop --headless from app_command and Fiji opens its GUI. The user does the work by hand and saves the result into outputs/; the block waits. If the user closes the app without producing output, the base run() raises BlockCancelledByAppError and the run is marked cancelled rather than failed.

What to look up

AppBlock, FileExchangeBridge, FileWatcher, validate_app_command, and the two exceptions live in scistudio.blocks.app — see the API reference for the full contract.