No Grammar to Rule Them All: A Survey of JSON-style DSLs for Visualization

Abstract

Augmenting text-based programming with rich structured interactions has been explored in many ways. Among these, projectional editors offer an enticing combination of structure editing and domain-specific program visualization. Yet such tools are typically bespoke and expensive to produce, leaving them inaccessible to many DSL and application designers.We describe a relatively inexpensive way to build rich projectional editors for a large class of DSLs-namely, those defined using JSON. Given any such JSON-based DSL, we derive a projectional editor through (i) a language-agnostic mapping from JSON Schemas to structure-editor GUIs and (ii) an API for application designers to implement custom views for the domainspecific types described in a schema. We implement these ideas in a prototype, Prong, which we illustrate with several examples including the Vega and Vega-Lite data visualization DSLs.

How Data Analysts Use a Visualization Grammar in Practice

“…GoG inspired many visualization grammars in multiple languages. McNutt provides an in-depth, literature-oriented survey and analysis of 57 visualization grammars (broadly defned as JSON-style DSLs) [38]. In practice, GoG-inspired grammars include ggplot2 [66] in R, Vega-Lite [55] in Javascript, Seaborn, 2 Altair (Vega-Lite frontend) [62], and plotnine 3 in Python, and Gadfy.jl, Algebra of Graphics in Julia.…”

“…Similar ecosystems exist in other languages. For example, JavaScript ecosystem contains JSON-based visualization grammars [38], data libraries like Arquero, 21 and Observable notebooks. Granted, the visualization grammars and systems on this continuum serve various purposes, and not every analyst can or need to specify arbitrary analysis through programming.…”

“…Since all our participants created visualizations with ggplot2, we assess the generalizability of our fndings in the context another GoG-based grammar, Vega-Lite [55]. Vega-Lite is similarly popular in practice (1.1M monthly downloads vs. ggplot2's 2.9M 23 ), and there is a vibrant research ecosystem around Vega-Lite [38,Figure 8].…”

“…One solution is to use visualization grammars, formalisms that create a wide range of visualizations by combinations of grammar components. Building from the original Grammar of Graphics (GoG) [70], visualization grammars have proliferated in the past decade (in literature as reviewed by Mc-Nutt [38] and in major scripting languages). Popular visualization grammars include ggplot2 [66] in the R language and the Vega ecosystem [55][56][57] in Javascript.…”

“…Despite the popularity and theoretical benefts of visualization grammars, we know little about whether or how analysts take advantage of these grammars in practice [38,49], especially given the potential tension between GoG design intention and analysts' needs. Understanding how analysts use visualization grammars can be crucial to improving grammar designs.…”