Nuclr FilePanel — Local Docker

This plugin presents the local Docker Engine as a normal Nuclr Commander source:

Docker
├── Containers
├── Images
└── Volumes

Containers and images are navigable, read-only filesystems. Files use the standard Nuclr viewer and F5 copies them into a local filesystem panel. Docker resource management is exposed through the normal panel function-key and context-menu action infrastructure.

Container lifecycle

Start, Stop, Restart, Pause and Resume act on the panel selection, not only on the row under the cursor: select several containers and one press applies the operation to all of them, exactly as Delete already behaves. Each operation is offered when at least one selected container is in a state that allows it, and is then applied only to the containers it is valid for — pressing Start on a half-running selection starts the stopped ones instead of raising an error for the rest. A container Docker refuses does not abandon the rest of the selection; the failures are collected and reported together once the progress dialog closes. Applying an operation to more than one container asks for confirmation first.

Function keys

The lifecycle commands sit on the Shift row because it is the only modifier row no desktop claims: macOS reserves Ctrl+F2–F8 system-wide and Linux window managers take Alt+F7/F8/F10, which is why Commander carries a stand-in chord for each of those rows and none for this one. Two Shift slots are still unavailable, so Resume does not continue the run past Pause: F9 is Commander's menu-bar key under every modifier and is never handed to a plugin, and Shift+F10 is the keyboard context-menu key on Windows, GTK and Qt alike. On macOS the function row needs Use F1, F2, etc. keys as standard function keys enabled, or fn held, as it does for every Commander plugin.

Integration design

• DockerService is the single Docker boundary. The first version uses Docker CLI commands with JSON output and argument vectors; it never parses human table output or builds shell strings. The executable can be overridden with the nuclr.docker.executable system property or the NUCLR_DOCKER_EXECUTABLE environment variable, for installations that are not on PATH.
• DockerFilesystemManager and TarDockerFilesystem provide one shared archive-like filesystem implementation for running/stopped containers and images. They do not use docker exec or require a shell or utilities inside an image.
• Image browsing creates a labelled, never-started temporary container, exports its merged filesystem, then removes the helper immediately. Helpers carry dev.nuclr.temporary=true and stale helpers older than 24 hours are cleaned during plugin initialization.
• Commander dispatches plugin actions on the event dispatch thread, so nothing that talks to Docker runs there: each action collects its input, then hands the work to DockerProgress, which runs it on a virtual thread behind a modal, cancellable dialog. Refresh is explicit; the plugin does not poll Docker.
• Remote and SSH Docker endpoints are rejected by this local-only implementation. The endpoint is re-read on refresh, so docker context use takes effect without restarting Commander.

Snapshot cache

A browsed container or image is exported once to a temporary tar and indexed in a single pass that records where each entry's bytes begin, so reading a file is a seek rather than another walk of the archive. Snapshots are bounded — four of them, or 8 GB, whichever comes first — and the least recently used one is dropped past that. Each export holds only a lock private to its own source, so a long image export never blocks browsing something else or unloading the plugin. Files left behind by a run that did not shut down cleanly are swept at startup.

Directory links are followed for display but not for descent: a copy or a recursive walk that reaches a directory linking back into itself — /usr/bin/X11 -> . ships in the Debian base image — copies it as an empty directory instead of repeating the tree.

Intentional first-version limits

• Docker filesystems are read-only. Only Docker-to-local copy is supported, and it replaces existing files after asking once, rather than prompting per file.
• Volume content browsing is deferred. Docker has no clean archive operation for a volume, and helper-container export excludes mounted volume data; the plugin does not pull a utility image or depend on shell tools merely to simulate browsing.
• Filesystem views are snapshots produced by docker container export. Refresh the panel to rebuild a snapshot. Docker export does not include mounted volume contents and some special pseudo-filesystems.
• Logs are a snapshot of the latest 10,000 lines rather than a live tail.

Tests

Run the normal Docker-independent suite with:

mvn test

It covers the pure helpers, the tar index and its link handling, the bounded snapshot cache, the copy and walk cycle guards, and the panel's own action behaviour through an in-memory DockerService. Opt into tests against a running local Docker daemon with NUCLR_DOCKER_TESTS=true.