Validating the Vega-Lite schema created by hvega

[DougBurke]

I've noticed, when viewing the output of hvega in the Vega-Lite editor that there are times we create invalid data. For instance, the following made-up example

vlShow (toVegaLite [ description "test"
                   , dataFromUrl "https://vega.github.io/vega-datasets/data/cars.json" []
                   , (encoding . position X [PName "Horsepower", PmType Quantitative,
                                             PScale [SReverse True]]
                               . position Y [PName "Miles_per_Gallon", PmType Quantitative]) []
                   , mark Circle []
                   ])

creates

{
  "mark": "circle",
  "data": {
    "url": "https://vega.github.io/vega-datasets/data/cars.json",
    "format": {"type": "json"}
  },
  "$schema": "https://vega.github.io/schema/vega-lite/v2.json",
  "encoding": {
    "x": {
      "field": "Horsepower",
      "scale": {"reverse": true},
      "type": "quantitative"
    },
    "y": {"field": "Miles_per_Gallon", "type": "quantitative"}
  },
  "description": "test"
}

and the Vega-Lite editor doesn't like the use of reverse, even though it does display the X axis in reverse.

[DougBurke]

In this particular case it's because SReverse leads to the creation of a Vega, rather than Vega-Lite, property (reverse). If you change

PScale [SReverse True]

to

PSort [Descending]

then the JSON passes the validator.

That's not to say we could do with more tests of this.

[DougBurke]

There are now tests, which generate a Vega-Lite schema and then compare it to an on-disk version (a "golden" file in the tasty vernacular). These can then be compared against the Vega-Lite schema as a check (and not all pass, particularly the examples using a custom projection).

It's not really sufficient, but is a start.