Command-Line Selection, Tag Expressions, Reruns, and Execution Profiles: Diagnostics, Failure Modes, and Production Practices

1. Production diagnostic sequence

1. Preserve original output.xml, log/report, command, and environment/profile evidence.
2. Confirm Python, Robot Framework, and optional RobotCode versions.
3. Confirm execution root, source path, selectors, argument files, variables, and test data revision.
4. Validate parse/import graph before changing search paths.
5. Inspect suite/test long names, tag membership, and selector combination.
6. Inspect library/external state only after confirming the intended tests actually ran.
7. If relevant, inspect timing, Pabot workers, CI workspace, and container paths.
8. Apply the least destructive correction.
9. Rerun the smallest controlled slice and keep both before/after evidence.

2. Failure mode: tag expression selects zero tests

# Intentionally broken assumption: no test has BOTH smoke and slow.
python -m robot --include smokeANDslow tests

Do not “fix” this with --runemptysuite in routine execution. First inspect tags and your intended predicate. If the real intent is “smoke OR slow,” use smokeORslow. If the intent is “smoke, but not slow,” use smokeNOTslow or include/exclude. The evidence is the selection count and exact expression, not intuition about what AND means.

3. Failure mode: simple name matches the wrong location

Large repositories often repeat names such as Health Check. A simple --test "Health Check" can match more than intended. Repair by using the full long name from the execution root or by combining a suite selector with the test selector.

python -m robot --suite "Tests.Service" --test "Health Check" tests
# Or when the long name is known:
python -m robot --test "Tests.Service.Health Check" tests

If you change the execution root, the long-name prefix changes. That is an identity change, not merely a path cosmetic.

4. Failure mode: argument file works only from one directory

# config/smoke.args contains a final line: tests
# From rf-cli-lab/ this is valid:
python -m robot -A config/smoke.args

# From the parent, this locates the argument file but the inserted `tests`
# argument now refers to the parent's cwd, not config/'s directory.
python -m robot -A rf-cli-lab/config/smoke.args

Repair the command contract, not Python import behavior. Options/data sources from argument files are inserted into the command-line argument stream. Use a documented project-root launcher, absolute path supplied by the launcher, or a wrapper that explicitly resolves project root. Arbitrary PYTHONPATH changes do not repair data-source cwd.

5. Failure mode: option/profile precedence surprise

Core argument-file options take precedence according to position: the file contents are inserted where -A appears. RobotCode adds a separate configuration-loading/profile layer. A local .robot.toml can therefore explain why two engineers using “smoke” get different output directories or variables.

# Core: reason about order explicitly.
python -m robot --outputdir results/a -A config/base.args --outputdir results/c tests

# Optional RobotCode diagnostics.
robotcode --version
robotcode profiles list
robotcode profiles show smoke

Do not assume RobotCode is active because a robot.toml exists; python -m robot does not load RobotCode profiles. Conversely, if you invoke RobotCode, inspect its resolved profile instead of debugging Robot Framework core for a wrapper setting.

6. Failure mode: rerun output replaces the wrong logical identity

Suppose the first run failed Tests.Service.Controlled Failure, then the file/suite is renamed to Tests.Service Checks.Controlled Failure before rerun. --rerunfailed uses prior failed identity against current source. The safest production rule is to rerun the same revision/configuration unless the change under verification is explicit and the mapping is understood.

7. Intentionally broken false-green workflow

# BROKEN POLICY — do not copy.
python -m robot --outputdir results/first tests || true
python -m robot --rerunfailed results/first/output.xml --outputdir results/rerun tests || true
python -m robot.rebot --merge --outputdir results/final \
  results/first/output.xml results/rerun/output.xml
# Then upload only results/final and delete results/first.  <-- evidence loss

This is broken for two reasons. Shell-level || true discards Robot's exit status, and deleting the first run destroys evidence of the original failure. A passing merged report may be a valid derived state, but it cannot be used to pretend the initial execution was clean.

Repair: capture each return code, preserve both raw outputs, run the rerun only under an explicit policy, merge into a third directory, and let the release gate distinguish first-pass success from pass-after-rerun.

8. Failure mode: include/exclude/skip semantics are conflated

If a command changes from --exclude quarantined to --skip quarantined, selected counts and report semantics change even though the quarantined test bodies still do not execute. That can be correct, but it is a policy change. Review the diff as an execution-contract change, not a cosmetic flag refactor.

9. Security: commands and profiles are evidence surfaces

Shell history, CI logs, argument files, robot.toml, and Robot log messages are not secret vaults. Never place a real token in --variable TOKEN:... in a copied command or committed file. Even Robot Framework 7.4 Secret values mask specific Robot surfaces; they do not encrypt files or guarantee that an external library will not log a value. Keep this chapter's values synthetic.

10. Diagnose the correct performance layer

11. Least-destructive rerun

After repairing selection, rerun the smallest slice that proves the correction. A zero-selection bug needs a selection-focused run, not the full suite. A profile output-dir bug can be diagnosed with RobotCode profile inspection plus one dry/simple test. Preserve the failing artifact before any new run.

Knowledge check

Summary and bridge

Selection failures are usually configuration/identity failures, not assertion failures. Lesson 5 combines the chapter into a checkpoint where you predict three slices, force one controlled failure, rerun/merge it, and produce an evidence packet plus precedence contract.

References and version anchors

• Robot Framework 7.4.2 User Guide — CLI option syntax, name/tag selection, argument files, reruns, skip/exclude semantics, and result post-processing.
• Robot Framework 7.4.2 on PyPI — stable release and Python compatibility anchor.
• RobotCode CLI reference — optional profile inspection and execution commands.