Due to the exploratory nature of computational notebook development, a notebook can be extensively evolved even though it is small, potentially incurring substantial technical debt. Indeed, in interview studies notebook authors have attested to performing ongoing tidying and big cleanups. However, many notebook authors are not trained as software developers, and environments like JupyterLab possess few features to aid notebook maintenance. As software refactoring is traditionally a critical tool for reducing technical debt, we sought to better understand the unique and growing ecology of computational notebooks by investigating the refactoring of public Jupyter notebooks. We randomly selected 15,000 Jupyter notebooks hosted on GitHub and studied 200 with meaningful commit histories. We found that notebook authors do refactor, favoring a few basic classic refactorings as well as those involving the notebook cell construct. Those with a computing background refactored differently than others, but not more so. Exploration-focused notebooks had a unique refactoring profile compared to more exposition-focused notebooks. Authors more often refactored their code as they went along, rather than deferring maintenance to big cleanups. These findings point to refactoring being intrinsic to notebook development.

We present a new technique for synthesizing dynamic, constraint-based visual layouts from examples. Our technique tackles two major challenges of layout synthesis. First, realistic layouts, especially on the web, often contain hundreds of elements, so the synthesizer needs to scale to layouts of this complexity. Second, in common usage scenarios, examples contain noise, so the synthesizer needs to be tolerant to imprecise inputs. To address these challenges we propose a two-phase approach to synthesis, where a local inference phase rapidly generates a set of likely candidate constraints that satisfy the given examples, and then a global inference phase selects a subset of the candidates that generalizes to unseen inputs. This separation of concerns helps our technique tackle the two challenges: the local phase employs Bayesian inference to handle noisy inputs, while the global phase leverages the hierarchical nature of complex layouts to decompose the global inference problem into inference of independent sub-layouts.