# Testing Notebook fails on Datasets

I have some functions that read in a .csv and convert them to a Dataset, and do some processing. I've written tests to validate that the correct transformations occur (so far very simple...like Identity). I execute the function inside of the testing notebook, it outputs a Dataset, and I convert the cell to a test.

When I execute the test, it Errors (not Fail). Mostly I see errors related to RecursionLimit for MakeBoxes. I've managed to replicate this with another basic dataset.

Dataset[<|"name" -> 1, "id" -> 2, "companyType" -> 3, "year" -> 6, "prerequisite" -> 8, "corequisite" -> 9, "cost" -> 10, "costPerUnitCotton" -> 11, "costPerUnitPoly" -> 12, "brandRep" -> 14, "vmRep" -> 15, "rmRep" -> 16, "chemRep" -> 17, "toxicsCotton" -> 18, "energyCottom" -> 19, "waterCotton" -> 20, "wasteCotton" -> 21, "toxicsPoly" -> 22, "energyPoly" -> 23, "waterPoly" -> 24, "wastePoly" -> 25|>]


Paste into a Testing notebook, execute, convert selection to test, and then run.

EDIT: forgot to mention that if I select a single element out of the Dataset so that it returns an Integer, the test runs successfully.

I'm using Mathematica 11.0.1.0 on Linux x86 64-bit, thanks!

• I pasted the Dataset into a Testing Notebook, highlighted the cell, clicked Convert Selection, then clicked Run. It failed (wrong results), as we would expect, since we didn't have an expected output. I see no error messages. This is the behavior I would expect. This was on 11.0.1 on Windows. Am I missing something? Or do you think it could be Linux only? – user6014 Nov 29 '16 at 14:41
• I executed the command to generate the result, and then converted selection, so there was an expected output. – Rick R Nov 29 '16 at 16:16
• Ah, yes. I see. It does appear as though this isn't being handled well in my environment either (11.0.1, Win10). Perhaps you should report this to support. – user6014 Nov 29 '16 at 16:21
• Thanks. I have reported this to support. Case 3782752. – Rick R Nov 29 '16 at 17:05
• Support responded, confirmed the issue, and is escalating to engineering. In the meantime they recommend that Normal be applied to datasets for testing. – Rick R Dec 2 '16 at 23:36