take into account slope at the end-points

[elmirjagudin]

Extrapolating the Y-value horizontally leads to false positives for curves with high derivative towards the end of the curve.

Take the reference curve's slope when calculating tube's end-points.

Solves https://github.com/modelica-tools/csv-compare/issues/13

[beutlich]

Thank you for this PR on #13. We evaluated it thoroughly and can give you the reasons for rejection now.

Explanation

The signal under test is outside of the tube for the end point. If the signal was continued with the same slope the end point would be inside the tube. It is neither known nor computable how signal shall be continued. If the signal was continued as constant, the compare decision is correct.

A convention needs to be found: Should false positive or false negative decisions be preferred? Since csv-compare is a tool for signal validation it should for doubtful cases rather issue an error (though there is none) instead of ignoring a real error.

Solution ideas

One idea is to add an option for the treatment of the end point. Then the risk of accepting a wrong signal as valid is transferred from the csv-compare tool to the user.

About the PR

• It is proposed to continue the tube under the assumption of a constant slope. This is not acceptable as general solution but would be as option.
• There is a risk of overflow, especially if the last two sample points occur at the same time (discontinuity).
• The algorithm ignores intermediate points along a straight line: This approach could also be used for the end point (by adding one point instead of two and replacing LY.Last() by reference.Y[reference.Count - 1] - size.Y and UY.Last() by reference.Y[reference.Count - 1] + size.Y).
• The comments ("add point top right", "false positive") are wrong.

Conclusion

This PR cannot be accepted in its current form.