Simplify syntax

[joelostblom]

As per the discussion in #2988 (reply in thread)

[mattijn]

One question. The text encoding channel is not included within the mark_bar().
Should the base not be just x and y and include the text as part of the mark_text().encode(text='wheat')?

[mcp292]

Personally, I thought that was the cool part: that you define your encoding and then the respective marks either use them in full, or in part.

[joelostblom]

To me, there is a conceptual difference between Altair and VegaLite here. In Altair, we can think of it as if it is the Chart object that takes the encoding, and then any number of marks can be added to it, as in the example in this PR:

base = alt.Chart(source).encode(
    x='wheat',
    y="year:O",
    text='wheat'
)
base.mark_bar() + base.mark_text(align='left', dx=1)

However, in VegaLite every mark has to be coupled to its own encoding (I think), so the reason the above spec works is that Altair expands it into a VegaLite spec that contains a redundant text encoding for the bars:

  "layer": [
    {
      "mark": {"type": "bar"},
      "encoding": {
        "text": {"field": "wheat", "type": "quantitative"},
        "x": {"field": "wheat", "type": "quantitative"},
        "y": {"field": "year", "type": "ordinal"}
      }
    },
    {
      "mark": {"type": "text", "align": "left", "dx": 2},
      "encoding": {
        "text": {"field": "wheat", "type": "quantitative"},
        "x": {"field": "wheat", "type": "quantitative"},
        "y": {"field": "year", "type": "ordinal"}
      }
    }
  ],

I believe that being able to think of an encoding as a property of a Chart and that any mark added to that chart can make use of those encodings is a useful level of abstraction to think about the Altair syntax even if it leads to slight increase in the VegaLite verbosity (notably, it is also similar to how users can think of their charts in ggplot and seaborns new api)

[joelostblom]

Actually, it seems like an encoding can belong to either a single layer or be shared across all layers in Vega-Lite too (by defining it at the Chart/Top level). That means that the conceptual thinking that I described above where an encoding belongs to the overall chart rather than a single mark can apply to Vega-Lite too but it depends on how the code is written (it always applies to Altair). I have broken this out to a separate issue in to discuss more #2999

[mattijn]

Thanks for the clarification @joelostblom, good that you opened another issue on this topic. Merging this!