# Opp Repl Chart Tests

> Detect visual regressions in result-analysis charts. run_chart_tests renders charts from current results and compares them pixel-wise against baseline images stored in the project's media_folder. Requires the `chart` extra (matplotlib + numpy). Load when analysis plots are part of your deliverable and must stay stable.

- Skill: `tabgab/opp-repl-chart-tests` (Agent Skill)
- Install (CLI): `npx skillmds@latest add tabgab/opp-repl-chart-tests`
- Raw SKILL.md: https://api.skillmd.com/api/skills/tabgab/opp-repl-chart-tests/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Coding & Dev Tools
- Author: tabgab (https://skillmd.com/u/tabgab)
- Updated: 2026-09-17
- Page: https://skillmd.com/skills/tabgab/opp-repl-chart-tests

---


# Chart tests

Chart tests render each analysis chart and compare the resulting
image against a baseline stored in the project's `media_folder`.
A visual diff catches regressions in analysis pipelines, not in the
simulation itself — e.g. a buggy pandas aggregation or a changed
axis label.

Upstream reference:
https://github.com/omnetpp/opp_repl/blob/main/doc/chart_tests.md

## Requirements

- `chart` extra installed: `pip install "opp_repl[chart]"`.
- `media_folder` set on the `SimulationProject` (INET convention:
  `doc/media`).

## Python API

    # Seed baselines (first time, or after intentional chart changes)
    update_chart_test_results(simulation_project=inet_project)

    # Run tests
    run_chart_tests(simulation_project=inet_project)

    # Scoped
    run_chart_tests(simulation_project=inet_project,
                    working_directory_filter="showcases")

Result codes: `PASS` / `FAIL` (`INSERT`/`UPDATE`/`KEEP` on updates).

## Command line

    opp_update_chart_test_results --load inet.opp -p inet
    opp_run_chart_tests           --load inet.opp -p inet

## When chart tests catch what fingerprint tests miss

- The numerical results are correct but the plotting code
  regressed.
- A new `matplotlib`/`seaborn` release changed default styles.
- An analysis notebook's filter criteria drifted.

## Pitfalls

- Matplotlib backends, fonts, and DPI all affect pixel output.
  Pin the stack (`matplotlib==X.Y`, optionally `MPLBACKEND=Agg`)
  across CI to avoid spurious diffs.
- Anti-aliasing quirks between Linux distros can cause flapping.
  For reproducible CI, run chart tests inside a fixed container.
- Tolerance is pixel-wise, not structural; a one-pixel shift may
  show as FAIL.  Consult the live docstring for tolerance knobs.
- When a chart test reports FAIL, inspect the pixel diff with
  `compare_charts()` in the REPL or the `opp_diff_charts` GUI
  (requires the `diffcharts` extra: `pip install "opp_repl[diffcharts]"`).

## See also

- `opp-repl-running-simulations` — produces the scalars charts read.
- `opp-repl-tasks-and-results` — inspect FAILed chart tests.
- `opp-repl-module-image-tests` — sibling suite for module/network
  *rendering* regressions (visual regression of how modules and
  networks are drawn), as opposed to the analysis-plot regressions
  caught here.