Turtles all the way down (0m13s)

Ashley Williams opens by reading the opening vignette from A Brief History of Time — the cosmologist, the heckler and the stack of cosmic tortoises — and signals that this will be a talk about ideas rather than code, with only two code slides in 40 minutes.

"the world is really a flat plate supported on the back of a giant tortoise"

"much like a Reddit comment, basically"

"this is a little bit more about the code for your computer up here"

Why even ask the question (3m50s)

She works at npm explaining how npm works, has been teaching beginners around the world, and noticed that two questions kept coming back at her: how do you get permission to publish a module, and the much trickier how do you decide what goes in a module.

"my job is to explain npm. So if you don't understand how npm works, that's my problem"

"teaching is nature's way of letting you know how sloppy your thinking is"

"have you ever really looked at your hands? But this is what we're going to do for the next 40 minutes"

What people actually said on Twitter (6m57s)

She tweeted "what's your primary motivation for writing or using modular code" and the answers clustered around maintainability, testability, reuse and separation of concerns — but when she pressed on how you decide what those single concerns are, almost everyone hedged or threw darts.

"I write so that my past self doesn't put my future self in a position where I'll need things I won't have"

"this person suggested that they just kind of used intuition... when I pressed this person they were like 'I throw darts, I hope other people have better strategies'"

"how many people had someone get mad at you when you just put it all in one file? All right, you're all liars"

Modularity is like time (10m05s)

The talk's theme: we use the concept of modularity constantly without being able to define it, in the same way we use "time" constantly and freeze when asked what it actually is.

"if you ask somebody hey what time is it, they'll look at their watch and they'll be like 3 p.m."

"if you ask 'well it's 3 p.m. but what is time?' you're going to get a response like 'I don't know'"

"this is why I have themed this talk with A Brief History of Time"

The npm-shaped universe and Dijkstra's hierarchies (11m10s)

A graph-database visualisation of npm — Lodash and React as glowing galaxies — gives the talk its scale, and Dijkstra supplies the reason: the conceptual hierarchies modern programming asks us to hold are deeper than anything in human history, and modularity is our coping mechanism.

"the npm registry has somewhere around 368,000 total packages... 6 billion downloads in the last month"

"the programmer is in the unique position [of having] to think in terms of conceptual hierarchies that are much deeper than a single mind ever needed to face before"

"the small modules philosophy... our short-term memory is finite"

Small modules and the left-pad moment (16m22s)

The argument that "left-pad means we've forgotten how to program" is dispatched as a no true Scotsman fallacy — we did not stop programming when we stopped writing assembly — and substack's framing is offered as the better rule: your application code should be the brackish residue that can't be abstracted away.

"abstraction is definitely an awesome power that we have as developers, but when we use it it doesn't mean that we forgot how to program"

"did we forget to program when we stopped writing C, when we stopped writing assembly?"

"when applications are done well they are just the really application-specific brackish residue that can't be so easily abstracted away"

The cost of small modules (25m30s)

Nolan Lawson's "the more I modularise my code the bigger it gets" puts the inverse case: webpack and Browserify preserve each module's scope at runtime, and that scope-lookup time costs end users real milliseconds. Rollup and Closure hoist everything into one function for that reason.

"alternative title: why one horse-sized JavaScript duck is faster than 100 duck-sized JavaScript horses"

"by preserving the modularity that was helping us in the development period, we actually are costing our users in runtime"

"it benefits the library writers, it disadvantages application writers, and it harms end users"

Software is change management; Parnas's real criterion (30m39s)

The deeper misconception is treating software as static. The 1972 Parnas paper she eventually found — On the criteria to be used in decomposing systems into modules — gives the answer Twitter could not: decompose along the seams of difficult decisions that are likely to change.

"the vast majority of software development is change management"

"nobody talked about their software changing, they all looked at it statically"

"start with a list of difficult design decisions, or design decisions which are likely to change; each module is then designed to hide such decisions from the others"

Code that is easy to delete (35m16s)

Tef's rule of thumb closes the loop: instead of writing reusable code, write disposable code — code that can be deleted cleanly — because deletion is the one operation that exposes whether your seams really are at the points of change.

"every line of code is written without reason, maintained out of weakness and deleted by chance"

"we should be writing code that is easy to delete, not easy to extend... instead of building reusable software we should try to build disposable software"

"often times when we're modularising things we will actually couple things more closely together, particularly if those things are rapidly changing"

SpaceTime, and a human with a brain (36m17s)

Ashley closes by reminding the audience that two observers in relative motion cannot agree on the order of events — let alone whether React is the best framework — so the only reliable tool we have is a human paying attention to the concepts that guide their decisions.

"we actually can't agree on space or time"

"modularity is an old problem, but in JavaScript we're dealing with it at a brand new scale and this scale is very hard to think about"

"right, this is why they pay a human with a brain who can investigate and learn"