That's it! Just replace let! with let_it_be. That's equal to the before_all approach but requires less refactoring.

That technique works pretty good but requires us to use instance variables and define everything at once. Thus it's not easy to refactor existing tests which use let/let! instead.

I've seen (and fixed) Ruby code that needed to be refactored for the client objects to use the accessor rather than the underlying mechanism, even though instance variables aren't directly visible. The underlying mechanism isn't always an instance variable - it can be delegations to or manipulations of a class you're hiding behind a facade, or a session store with a particular format, or all kinds. And it can change. 'Self-encapsulation' can help if you need to swap a technology, a library, an object specification, etc.

Also, Sandi Metz mentions this in POODR. As I recall, she also advocates wrapping bare instance variables in methods, even when they're only used internally. It helps avoid mad refactoring later.

But sure, go ahead and enforce self-encapsulation if you like; it makes it easier to do memoization or whatever later on.

The number one problem that I see developers have when practicing test-first development that impedes them from refactoring their code is that they over-specify behavior in their tests. This leads developers to write more tests than are needed, which can become a burden when refactoring code.

Unfortunately, given how widely used concat_javascript_sources is, this required changing a lot of tests. It would be nice if we could remove some of the duplication in these tests (so that similar changes would not require updating this many tests), but that can come in another PR.

Jbuilder gives you a simple DSL for declaring JSON structures that beats manipulating giant hash structures. This is particularly helpful when the generation process is fraught with conditionals and loops.

hyperscript is much simpler to refactor and DRY up your code than with JSX, because, being vanilla javascript, its easier to work with variable assignment, loops and conditionals.

Templates are prone to unnoticed runtime errors, are hard to test, and are not easy to restructure or decompose.

When a component reaches such a size that this becomes a problem, the obvious course of action is to refactor it into multiple components. But the refactoring is complex for the same reason: extracting the styles that relate to a particular piece of markup is an error-prone manual process, where the relevant styles may be interleaved with irrelevant ones.

(At the point at which it does make sense to turn this into a separate Tooltip.svelte component, the extraction is a completely mechanical process that could even be automated by tooling.)