unidecompiler-gui

unidecompiler-gui is a read-only PySide6 workbench for the public unidecompiler.DecompilerEngine API. It accepts one artifact, a directory, or a ZIP/JAR archive and shows pseudocode, AST, bytecode, and diagnostics together.

Install from PyPI; cloning this repository is not required for normal use.

Install the base GUI with its Qt dependency:

python -m pip install unidecompiler-gui

Install all separately published frontend plugins when needed:

python -m pip install 'unidecompiler-gui[all-formats]'

The GUI never imports frontend plugin packages directly. It discovers installed plugins through DecompilerEngine and does not modify input artifacts or save workspace state. Its optional Simulation tab uses the separate generic IR simulator and can load a trusted Python runtime file for unresolved functions.

Frontend persistence

Custom VM frontend folders registered from View -> Frontends are persisted by the GUI host. After the first successful registration, the GUI restores the folder on the next startup before opening input files. The frontend registry and core remain runtime-only; the GUI stores only the normalized source path and frontend ID.

Unload selected removes a frontend from the current process but keeps its startup record. Remove from startup removes it from the current process and from the GUI startup records, but never deletes the frontend directory from disk. Built-in entry-point frontends are not stored as user records.

If a saved directory is missing or its manifest/module is invalid, startup continues. The frontend manager marks the record unavailable and shows the diagnostic so the path can be repaired or removed. Frontend source is trusted Python code and is not sandboxed.

GUI plugins

The Plugins menu manages optional Python GUI plugins. A plugin is an application-layer extension, separate from bytecode frontend plugins. It can inspect immutable snapshots of open documents, functions, AST/reference summaries and selections; add commands and declarative panels; request navigation through stable IDs/source locations; and start asynchronous simulation jobs.

Plugins cannot modify artifacts, generic IR, AST, pseudocode, frontend registration, or simulator execution. They never receive a Qt Workbench, frontend decoder payload, ModuleIR/FunctionIR, simulator frame, or stack. Function lookup remains frontend-owned: a plugin submits an opaque query while the GUI host delegates execution to the generic simulator.

Plugins are trusted in-process Python code, with the same permissions as the GUI process. Review every local folder or GitHub repository before installing. Dependencies in the manifest are checked at load time but never installed automatically. Install, update, enable, disable, and removal take effect after restart.

Plugin layout

Install a folder with plugin.toml through Plugins -> Manage plugins, or install a GitHub owner/repository or /tree/ref URL.

[plugin]
id = "example.function-browser"
name = "Function browser"
version = "1.0.0"
api = "1"
entry = "function_browser:register"

[python]
requires = []

from unidecompiler_gui_sdk import Command, Panel, PanelState

def register(context):
    context.panels.register(Panel("functions", "Functions"))

    def refresh(ctx):
        document = ctx.active_document
        rows = () if document is None else tuple((item.name, item.status) for item in document.functions)
        ctx.set_panel_state("functions", PanelState.table(("Function", "Status"), rows))

    context.commands.register(Command("refresh", "Refresh function browser", refresh))
    context.subscribe("document_selected", lambda _document: refresh(context))

Use only unidecompiler_gui_sdk types. Its API is Qt-neutral and versioned; plugin.toml must declare the matching API version. The repository includes unidecompiler-gui-test-plugin/ as a complete working example.

To simulate, first call context.request_simulation_targets(document_id). It returns immediately with a target-discovery job; observe simulation_targets_completed, obtain its SimulationTargetSnapshot values through context.get_target_job(job_id), then pass a selected snapshot's opaque query to context.submit_simulation(...). Observe simulation_completed and use the SDK's SimulationResultSnapshot; plugins never receive simulator runner objects or frontend adapters.