coerce to the scale’s type (#532)

* coerce to the scale’s type

* upgrade isoformat

* document type coercion

* language

* changelog

* update CHANGELOG

* coerce invalid dates to undefined

* update README

* update README

Co-authored-by: Philippe Rivière <[email protected]>

Author: mbostock

CHANGELOG.md

@@ -1,5 +1,13 @@
 # Observable Plot - Changelog
 
+## 0.3.0
+
+*Not yet released.* These notes are a work in progress.
+
+### Scales
+
+Quantitative scales, as well as identity position scales, now coerce channel values to numbers; both null and undefined are coerced to NaN. Similarly, time scales coerce channel values to dates; numbers are assumed to be milliseconds since UNIX epoch, while strings are assumed to be in [ISO 8601 format](https://github.com/mbostock/isoformat/blob/main/README.md#parsedate-fallback).
+
 ## 0.2.0
 
 Released August 20, 2021.

README.md

@@ -165,6 +165,8 @@ For ordinal data (*e.g.*, strings), use the *ordinal* scale type or the *point*
 
 You can opt-out of a scale using the *identity* scale type. This is useful if you wish to specify literal colors or pixel positions within a mark channel rather than relying on the scale to convert abstract values into visual values. For position scales (*x* and *y*), an *identity* scale is still quantitative and may produce an axis, yet unlike a *linear* scale the domain and range are fixed based on the plot layout.
 
+Quantitative scales, as well as identity position scales, coerce channel values to numbers; both null and undefined are coerced to NaN. Similarly, time scales coerce channel values to dates; numbers are assumed to be milliseconds since UNIX epoch, while strings are assumed to be in [ISO 8601 format](https://github.com/mbostock/isoformat/blob/main/README.md#parsedate-fallback).
+
 A scale’s domain (the extent of its inputs, abstract values) and range (the extent of its outputs, visual values) are typically inferred automatically. You can set them explicitly using these options:
 
 * *scale*.**domain** - typically [*min*, *max*], or an array of ordinal or categorical values
@@ -1532,7 +1534,7 @@ These helper functions are provided for use as a *scale*.tickFormat [axis option
 Plot.formatIsoDate(new Date("2020-01-01T00:00.000Z")) // "2020-01-01"
 ```
 
-Given a *date*, returns the shortest equivalent ISO 8601 UTC string.
+Given a *date*, returns the shortest equivalent ISO 8601 UTC string. If the given *date* is not valid, returns `"Invalid Date"`.
 
 #### Plot.formatWeekday(*locale*, *format*)
 

package.json

@@ -51,7 +51,7 @@
   },
   "dependencies": {
     "d3": "^7.0.0",
-    "isoformat": "^0.1.0"
+    "isoformat": "^0.2.0"
   },
   "engines": {
     "node": ">=12"

src/format.js

@@ -1,4 +1,4 @@
-export {default as formatIsoDate} from "isoformat";
+import {format as isoFormat} from "isoformat";
 
 export function formatMonth(locale = "en-US", month = "short") {
   const format = new Intl.DateTimeFormat(locale, {timeZone: "UTC", month});
@@ -17,3 +17,7 @@ export function formatWeekday(locale = "en-US", weekday = "short") {
     }
   };
 }
+
+export function formatIsoDate(date) {
+  return isoFormat(date, "Invalid Date");
+}

src/scales.js

@@ -4,6 +4,7 @@ import {ScaleDiverging, ScaleDivergingSqrt, ScaleDivergingPow, ScaleDivergingLog
 import {ScaleTime, ScaleUtc} from "./scales/temporal.js";
 import {ScaleOrdinal, ScalePoint, ScaleBand} from "./scales/ordinal.js";
 import {isOrdinal, isTemporal} from "./mark.js";
+import {parse as isoParse} from "isoformat";
 
 export function Scales(channels, {inset, round, nice, align, padding, ...options} = {}) {
   const scales = {};
@@ -58,7 +59,37 @@ function autoScaleRound(scale) {
 }
 
 function Scale(key, channels = [], options = {}) {
-  switch (inferScaleType(key, channels, options)) {
+  const type = inferScaleType(key, channels, options);
+
+  // Once the scale type is known, coerce the associated channel values and any
+  // explicitly-specified domain to the expected type.
+  switch (type) {
+    case "diverging":
+    case "diverging-sqrt":
+    case "diverging-pow":
+    case "diverging-log":
+    case "diverging-symlog":
+    case "cyclical":
+    case "sequential":
+    case "linear":
+    case "sqrt":
+    case "threshold":
+    case "quantile":
+    case "pow":
+    case "log":
+    case "symlog":
+      options = coerceType(channels, options, coerceNumber, Float64Array);
+      break;
+    case "identity":
+      if (registry.get(key) === position) options = coerceType(channels, options, coerceNumber, Float64Array);
+      break;
+    case "utc":
+    case "time":
+      options = coerceType(channels, options, coerceDate);
+      break;
+  }
+
+  switch (type) {
     case "diverging": return ScaleDiverging(key, channels, options);
     case "diverging-sqrt": return ScaleDivergingSqrt(key, channels, options);
     case "diverging-pow": return ScaleDivergingPow(key, channels, options);
@@ -144,3 +175,33 @@ export function isCollapsed(scale) {
   }
   return true;
 }
+
+// Mutates channel.value!
+function coerceType(channels, options, coerce, type) {
+  for (const c of channels) c.value = coerceArray(c.value, coerce, type);
+  return {...options, domain: coerceArray(options.domain, coerce, type)};
+}
+
+function coerceArray(array, coerce, type = Array) {
+  if (array !== undefined) return type.from(array, coerce);
+}
+
+// Unlike Mark’s number, here we want to convert null and undefined to NaN,
+// since the result will be stored in a Float64Array and we don’t want null to
+// be coerced to zero.
+function coerceNumber(x) {
+  return x == null ? NaN : +x;
+}
+
+// When coercing strings to dates, we only want to allow the ISO 8601 format
+// since the built-in string parsing of the Date constructor varies across
+// browsers. (In the future, this could be made more liberal if desired, though
+// it is still generally preferable to do date parsing yourself explicitly,
+// rather than rely on Plot.) Any non-string values are coerced to number first
+// and treated as milliseconds since UNIX epoch.
+function coerceDate(x) {
+  return x instanceof Date && !isNaN(x) ? x
+    : typeof x === "string" ? isoParse(x)
+    : x == null || isNaN(x = +x) ? undefined
+    : new Date(x);
+}

test/output/aaplCloseUntyped.svg

@@ -0,0 +1,19 @@
+import * as Plot from "@observablehq/plot";
+import * as d3 from "d3";
+
+export default async function() {
+  const AAPL = await d3.csv("data/aapl.csv");
+  return Plot.plot({
+    x: {
+      type: "utc"
+    },
+    y: {
+      type: "linear",
+      grid: true
+    },
+    marks: [
+      Plot.line(AAPL, {x: "Date", y: "Close"}),
+      Plot.ruleY([0])
+    ]
+  });
+}

test/plots/aapl-close-untyped.js

@@ -1,6 +1,7 @@
 export {default as aaplCandlestick} from "./aapl-candlestick.js";
 export {default as aaplChangeVolume} from "./aapl-change-volume.js";
 export {default as aaplClose} from "./aapl-close.js";
+export {default as aaplCloseUntyped} from "./aapl-close-untyped.js";
 export {default as aaplMonthly} from "./aapl-monthly.js";
 export {default as aaplVolume} from "./aapl-volume.js";
 export {default as anscombeQuartet} from "./anscombe-quartet.js";

test/plots/index.js

@@ -2243,10 +2243,10 @@ isexe@^2.0.0:
   resolved "https://registry.yarnpkg.com/isexe/-/isexe-2.0.0.tgz#e8fbf374dc556ff8947a10dcb0572d633f2cfa10"
   integrity sha1-6PvzdNxVb/iUehDcsFctYz8s+hA=
 
-isoformat@^0.1.0:
-  version "0.1.0"
-  resolved "https://registry.yarnpkg.com/isoformat/-/isoformat-0.1.0.tgz#b693c1c9ee9ab02f1af5af41ceeae52bf501b233"
-  integrity sha512-4wCSk50Ov1PKbZ2m+YN0rUgQfF4NRkIavbhpW1mANEqD9HxBZ+j/fWk8hERq1yxn+CfWqvOac4m9axLuF0NfEw==
+isoformat@^0.2.0:
+  version "0.2.0"
+  resolved "https://registry.yarnpkg.com/isoformat/-/isoformat-0.2.0.tgz#52c3dce6c281adb6cb7f060895a731b7b2d52c1b"
+  integrity sha512-iyxQ94xMvUZryoHVaXg/TSLM318/aO7xS7Ute+t4MkvZ17IDfe9MkI/MQuu7XgxbmTiGkeggNj+1f6wmxF876Q==
 
 isstream@~0.1.2:
   version "0.1.2"