d3-format@1.4.2 sideEffects:false breaks graph axis formatting

[dpmott]

d3-format@1.4.2 sideEffects:false breaks graph axis formatting

The d3-format src\defaultLocale.js calls defaultLocale at load time. This is, by definition, a side effect of loading the module. As such, it's inappropriate to set sideEffects:false in the package.json file. I believe that this is the root cause of this issue.

For our project compiled for production, d3-format@1.4.2 causes d3-scale\src\tickFormat.js to exception when it calls format(), because format has not been defined, presumably because webpack's tree shaking has entirely removed d3-format's invocation of defaultLocale().

The docs at https://www.npmjs.com/package/d3-format say this:

If you do not set a default locale, it defaults to U.S. English.

However, this is not true when compiling in production mode with d3-format@1.4.2.

When I explicitly install d3-format@1.4.1, or if I explicitly call d3d3.formatDefaultLocale(), then I do not get an exception, and my graphs render in my application.

[mbostock]

My understanding is that sideEffects: false is intended to mean there are no side-effects that cross modules, not that there are no side-effects within the module. That d3-format has an internal side-effect does not violate the intent of “sideEffects: false”.

What are you using to compile your code?

[dpmott]

We're using stock Angular 7.2.15, typescript 3.2.4, node v10.16.3, npm 6.9.0.

The failure comes in this (pseudo)code:

private svg: d3.Selection<SVGGElement, number, HTMLElement, number>;
private xScale: d3.ScaleLinear<number, number>;
private xAxis: d3.Axis<number | { valueOf(): number }>;
protected selector: string; // set to something like '#my-id .my-chart'

this.svg = d3.select(this.selector)
      .append('svg')
      .attr('width', ...)
      .attr('height', ...)
      .append('g')
      .attr('transform', `translate(0, ...)`);

this.xScale = d3.scaleLinear()...

this.xAxis = d3.axisBottom(this.xScale)
      .ticks(...)
      .tickValues(...)
      .tickSize(...);

this.svg.append('g')
      .attr('class', 'x-axis')
      .attr('transform', `translate(0, ...)`)
      .call(this.xAxis) // <--- this exceptions in a production build

The error is:

TypeError: (void 0) is not a function
    at tn (vendor.ed9171441c8bdb36a768.js:1)
    at Function.e.tickFormat (vendor.ed9171441c8bdb36a768.js:1)
    at d (vendor.ed9171441c8bdb36a768.js:1)
    at C.call (vendor.ed9171441c8bdb36a768.js:1)
    at t.drawXAxis (main.6d696a7192e551f1463d.js:1)
    at t.render (main.6d696a7192e551f1463d.js:1)
    at n.ngAfterViewChecked (main.6d696a7192e551f1463d.js:1)
    at Xi (vendor.ed9171441c8bdb36a768.js:1)
    at $i (vendor.ed9171441c8bdb36a768.js:1)
    at Qi (vendor.ed9171441c8bdb36a768.js:1)

I instrumented various functions in the d3 ecosystem, and determined that defaultLocale() in d3-format src\defaultLocale.js was never getting called in a production build. When I added a call from our code to d3.formatDefaultLocale() (which is the exported name for d3-format's defaultLocale()), the above call to .call(this.xAxis) completed without error.

The only production change that I see between d3-format v1.4.1 and v1.4.2 is the addition of the sideEffects:false flag in package.json. 🤷‍♂

[dpmott]

Here's a link to the webpack docs that may help:

https://webpack.js.org/guides/tree-shaking/#mark-the-file-as-side-effect-free

To quote:

A "side effect" is defined as code that performs a special behavior when imported, other than exposing one or more exports. An example of this are polyfills, which affect the global scope and usually do not provide an export.

I would think that some global variable initialization would meet this definition of "side effect".

[mbostock]

Angular’s optimizer is known to break code, and they don’t appear to have plans to fix this behavior. This has affected other D3 libraries (e.g., d3/d3-color#68 (comment)), and may be the cause of what’s happening here.

This change to add sideEffects: false was suggested by a Webpack maintainer; see d3/d3#3131. I don’t personally use Webpack, and I find the documentation to be somewhat ambiguous (does a “module” mean any individual file, as in ES modules, or do they mean the package in total, as in npm & node_modules?). I don’t know how to determine what is appropriate here.

From my perspective, the only thing adding sideEffects: false has brought is an influx of issues from Angular users. (Sorry for the gripe!)

[mbostock]

Proposed fix in #85. I think we’ll need a similar fix for d3-time-format.

[mbostock]

No response, but I merged #85, so I’ll publish that and see what happens.

[dpmott]

Thanks for the quick resolution on this, and Happy New Year! ❤️