Should time series be allowed to be modified if they are used by an existing stored model?

[dbrakenhoff]

Some points to discuss:

• Appending new measurements to an existing time series should be allowed, regardless if it is used by a model or not
• Deleting an oseries that is used by an existing model should raise an error? (maybe already does?)
• Updating existing observations should raise an error if oseries/stresses are used by an existing model?
• Errors can be ignored by specifying some keyword argument

[OnnoEbbens]

• Appending new measurements to an existing time series should be allowed, regardless if it is used by a model or not

Yes but it should raise an (ignorable) error if the time series is used by a model (see 3rd point)

• Deleting an oseries that is used by an existing model should raise an error? (maybe already does?)

Yes. If you ignore the error the model should be removed from the pastastore

• Updating existing observations should raise an error if oseries/stresses are used by an existing model?

Yes. If you ignore the error the solved model should be brought back to its unsolved state. I think you can save an unsolved model in the pastastore right?

• Errors can be ignored by specifying some keyword argument

Yes, I like how pandas deals with this. For example in the pd.to_numeric function you have the keyword argument errors which can be 'raise' to raise an error, 'coerce' which tries to do it anyway or 'ignore' which ignores the error. We could do with two options I guess 'error' as default and 'coerce' to modify a time series and unsolve all related models. I see now that pandas will deprecate the 'ignore' option in newer versions in favor of explicit error handling.