Should we run some basic checks after parsing a model?

[MichaelClerx]

At the moment we have some methods to check the model makes sense:

• Model.check_left_right_units_equal

and

• testparser.py::check_rdf_identities(model)
• testparser.py::check_dummy_assignment(model)

(I moved these last two into testing at one stage, as that was the only place they were used)

Should these be standard checks that occur after parsing a model? And maybe all hidden but called by some validate() method?

• Add a (validation) check that a model has only one free variable

[MichaelClerx]

Thoughts @jonc125 @MauriceHendrix @skeating ?

[MauriceHendrix]

It's not a bad idea but I do want to be a bit careful.

Depending on the validation my tests might be harder (probably not but I'd want to test it first)
Also: will there be a performance implication?
Is there a chance of disallowing something that's perfectly legal (like unrelated metadata etc)?

In any case I'd prefer to have an option to bypass the validation if I felt like it.

[MichaelClerx]

I'm guessing there won't be much of a performance hit

But we could definitely make the validation optional. Have some kind validate method that returns True or False.

I'd want that anyway for the use case where users build a model from scratch through the API, or when they modify a model after loading it

[MauriceHendrix]

I'm guessing there won't be much of a performance hit

But we could definitely make the validation optional. Have some kind validate method that returns True or False.

I'd want that anyway for the use case where users build a model from scratch through the API, or when they modify a model after loading it

I don't might if the deaful is on for the validation

[jonc125]

Happy to start curating a list of possible checks, but let's not implement any yet. Some like thorough units checking would fail for most existing models, for instance... so we need to be selective!

[skeating]

I agree this is important. Experience from libSBML would suggest having categories of tests that can be turned on/off. Some people just really dont care if something is not quite right but their model still runs and would get frustrated by a model not being read in just because of an error that they dont see as a problem.

[MauriceHendrix]

I guess that's what warnings are for?

[skeating]

Yes but other people who want things to be strict and fail completely if there are discrepancies. Modellers come in all shapes and sizes 😄

[MichaelClerx]

Also if you're using cellmlmanip you wouldnt' want to filter warnings to see which ones to keep as warnings and which ones to show as an error to the user

[MauriceHendrix]

Yes but other people who want things to be strict and fail completely if there are discrepancies. Modellers come in all shapes and sizes 😄

ha ha sure that's fair enough