RobotCode Files

Features

• robot: Make the analysis cache trustworthy under rapid file changes (c998399)

Fast consecutive changes to several files could poison the analysis cache: the editor and the CLI then reported errors that did not exist, like missing keywords or unresolvable variables, until the affected files were touched again. Cached results are now bound to exactly the file state they were computed from, and unsaved editor content is analyzed but never cached, so stale results can no longer stick.

• robotcode: Run tests through a wrapper (7aa84f6)

Run the test execution through a command of your choosing, so you can prepare the environment a run needs - start the services, servers, mocks or session the tests depend on, run the tests inside it, and tear it down again afterwards. For anything that is only environment variables, keep using robot.toml's [env].

Configure it per profile in robot.toml:

  [profiles.integration]
  wrapper = ["./with-test-services.sh"]

set it ad hoc with --wrapper (or the ROBOTCODE_WRAPPER env var), and turn a configured wrapper off for one run with --no-wrapper. It applies to every command that executes Robot Framework - run, debug, the REPL and the notebook kernel - never to discovery, the language server or libdoc.

Editors pick it up automatically because they run robotcode; the VS Code setting robotcode.debug.launchWrapper can override it for the IDE alone.

Bug Fixes

• plugin: Use compact JSON separators for json output format (e882368)

The compact flag was compared against OutputFormat.TEXT instead of OutputFormat.JSON, so compact JSON output was never actually applied.

• robotcode: Make the wrapper work on Windows (7baa0ac)

The wrapper re-run used os.execvp to replace the process, which only has POSIX semantics; on Windows os.exec* spawns a new process and exits the caller, so the exit code and stdout/stderr don't propagate. Keep execvp on POSIX (clean, no extra process) and on Windows spawn the command, wait, and forward its exit code. The subprocess tests, and their wrapper scripts, use spawn-and-wait too, so they no longer skip on Windows.

Also document how running through a wrapper differs on Linux/macOS vs. Windows, and why.

• robotcode: Make the wrapper work in VS Code (1180d9c)

Running or debugging tests through a wrapper did not work from the editor, where robotcode runs from a bundled path instead of being installed in the interpreter. Two things are fixed:

• The debug launcher no longer applies the wrapper itself; it forwards a launch.json wrapper to robotcode as --wrapper and lets robotcode apply it (so a launch.json wrapper still overrides the profile's).

• When robotcode re-executes itself under the wrapper it re-runs through the same entry it was started from (--launcher-script) instead of python -m robotcode.cli, which the bundled interpreter cannot import. ROBOTCODE_BUNDLED_ROBOTCODE_MAIN is deliberately not used for this: VS Code sets it in every terminal, so it would divert a directly started robotcode to the bundled copy.

• vscode: Adapt to vscode-languageclient 10 type changes (95f305a)

Bumping vscode-languageclient to 10.1.0 broke the typecheck: LanguageClient now requires a LogOutputChannel instead of a plain OutputChannel, and Diagnostic.message can now be a MarkupContent instead of just a string.

• vscode: Restore -I and -N options when running tests (c76b113)

Running a single test, a file, or any subset from the Test Explorer again passes Robot Framework's parse-include (-I) and top-level suite name (-N) options. Since 2.6.0 both were missing, so on Robot Framework 6.1+ the run parsed the whole workspace and surfaced parse errors from unrelated files.

Robot Framework's --parseinclude support is now reported as a per-project fact on the discovery result envelope. In the robotcode discover -f json output this is the new supportsParseInclude key on the result (replacing the previous needsParseInclude key on each item).

Documentation

• news: Add v2.7.0 release notes (033e12d)

Announce the test-run wrapper, the trustworthy analysis cache, and the restored -I/-N options for Test Explorer subset runs.