Improve data type analyses for arbitrary values (#9320)

* improve split logic by delimiter

The original RegEx did mostly what we want, the idea is that we wanted
to split by a `,` but one that was not within `()`. This is useful when
you define multiple background colors for example:
```html
<div class="bg-[rgb(0,0,0),rgb(255,255,255)]"></div>
```

In this case splitting by the regex would result in the proper result:
```js
let result = [
  'rgb(0,0,0)',
  'rgb(255,255,255)'
]
```

Visually, you can think of it like:
```
    ┌─[./example.html]
    │
∙ 1 │   <div class="bg-[rgb(0,0,0),rgb(255,255,255)]"></div>
    ·                       ──┬── ┬    ─────┬─────
    ·                         │   │         ╰─────── Guarded by parens
    ·                         │   ╰───────────────── We will split here
    ·                         ╰───────────────────── Guarded by parens
    │
    └─
```

We properly split by `,` not inside a `()`. However, this RegEx fails
the moment you have deeply nested RegEx values.

Visually, this is what's happening:
```
    ┌─[./example.html]
    │
∙ 1 │   <div class="bg-[rgba(0,0,0,var(--alpha))]"></div>
    ·                         ┬ ┬ ┬
    ·                         ╰─┴─┴── We accidentally split here
    │
    └─
```
This is because on the right of the `,`, the first paren is an opening
paren `(` instead of a closing one `)`.

I'm not 100% sure how we can improve the RegEx to handle that case as
well, instead I wrote a small `splitBy` function that allows you to
split the string by a character (just like you could do before) but
ignores the ones inside the given exceptions. This keeps track of a
stack to know whether we are within parens or not.

Visually, the fix looks like this:
```
    ┌─[./example.html]
    │
∙ 1 │   <div class="bg-[rgba(0,0,0,var(--alpha)),rgb(255,255,255,var(--alpha))]"></div>
    ·                         ┬ ┬ ┬             ┬       ┬   ┬   ┬
    ·                         │ │ │             │       ╰───┴───┴── Guarded by parens
    ·                         │ │ │             ╰────────────────── We will split here
    ·                         ╰─┴─┴──────────────────────────────── Guarded by parens
    │
    └─
```

* use already existing `splitAtTopLevelOnly` function

* add faster implemetation for `splitAtTopLevelOnly`

However, the faster version can't handle separators with multiple
characters right now. So instead of using buggy code or only using the
"slower" code, we've added a fast path where we use the faster code
wherever we can.

* use `splitAtTopLevelOnly` directly

* make split go brrrrrrr

* update changelog

* remove unncessary array.from call

Co-authored-by: Jordan Pittman <jordan@cryptica.me>

Author: RobinMalfait

CHANGELOG.md

@@ -31,6 +31,7 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 - Don't mutate shared config objects ([#9294](https://github.com/tailwindlabs/tailwindcss/pull/9294))
 - Fix ordering of parallel variants ([#9282](https://github.com/tailwindlabs/tailwindcss/pull/9282))
 - Handle variants in utility selectors using `:where()` and `:has()` ([#9309](https://github.com/tailwindlabs/tailwindcss/pull/9309))
+- Improve data type analyses for arbitrary values ([#9320](https://github.com/tailwindlabs/tailwindcss/pull/9320))
 
 ## [3.1.8] - 2022-08-05
 

src/lib/generateRules.js

@@ -461,7 +461,7 @@ function splitWithSeparator(input, separator) {
     return [sharedState.NOT_ON_DEMAND]
   }
 
-  return Array.from(splitAtTopLevelOnly(input, separator))
+  return splitAtTopLevelOnly(input, separator)
 }
 
 function* recordCandidates(matches, classCandidate) {

src/util/dataTypes.js

@@ -1,13 +1,11 @@
 import { parseColor } from './color'
 import { parseBoxShadowValue } from './parseBoxShadowValue'
+import { splitAtTopLevelOnly } from './splitAtTopLevelOnly'
 
 let cssFunctions = ['min', 'max', 'clamp', 'calc']
 
 // Ref: https://developer.mozilla.org/en-US/docs/Web/CSS/CSS_Types
 
-let COMMA = /,(?![^(]*\))/g // Comma separator that is not located between brackets. E.g.: `cubiz-bezier(a, b, c)` these don't count.
-let UNDERSCORE = /_(?![^(]*\))/g // Underscore separator that is not located between brackets. E.g.: `rgba(255,_255,_255)_black` these don't count.
-
 // This is not a data type, but rather a function that can normalize the
 // correct values.
 export function normalize(value, isRoot = true) {
@@ -61,7 +59,7 @@ export function number(value) {
 }
 
 export function percentage(value) {
-  return value.split(UNDERSCORE).every((part) => {
+  return splitAtTopLevelOnly(value, '_').every((part) => {
     return /%$/g.test(part) || cssFunctions.some((fn) => new RegExp(`^${fn}\\(.+?%`).test(part))
   })
 }
@@ -86,7 +84,7 @@ let lengthUnits = [
 ]
 let lengthUnitsPattern = `(?:${lengthUnits.join('|')})`
 export function length(value) {
-  return value.split(UNDERSCORE).every((part) => {
+  return splitAtTopLevelOnly(value, '_').every((part) => {
     return (
       part === '0' ||
       new RegExp(`${lengthUnitsPattern}$`).test(part) ||
@@ -115,7 +113,7 @@ export function shadow(value) {
 export function color(value) {
   let colors = 0
 
-  let result = value.split(UNDERSCORE).every((part) => {
+  let result = splitAtTopLevelOnly(value, '_').every((part) => {
     part = normalize(part)
 
     if (part.startsWith('var(')) return true
@@ -130,7 +128,7 @@ export function color(value) {
 
 export function image(value) {
   let images = 0
-  let result = value.split(COMMA).every((part) => {
+  let result = splitAtTopLevelOnly(value, ',').every((part) => {
     part = normalize(part)
 
     if (part.startsWith('var(')) return true
@@ -171,7 +169,7 @@ export function gradient(value) {
 let validPositions = new Set(['center', 'top', 'right', 'bottom', 'left'])
 export function position(value) {
   let positions = 0
-  let result = value.split(UNDERSCORE).every((part) => {
+  let result = splitAtTopLevelOnly(value, '_').every((part) => {
     part = normalize(part)
 
     if (part.startsWith('var(')) return true
@@ -189,7 +187,7 @@ export function position(value) {
 
 export function familyName(value) {
   let fonts = 0
-  let result = value.split(COMMA).every((part) => {
+  let result = splitAtTopLevelOnly(value, ',').every((part) => {
     part = normalize(part)
 
     if (part.startsWith('var(')) return true

src/util/parseBoxShadowValue.js

@@ -5,7 +5,7 @@ let SPACE = /\ +(?![^(]*\))/g // Similar to the one above, but with spaces inste
 let LENGTH = /^-?(\d+|\.\d+)(.*?)$/g
 
 export function parseBoxShadowValue(input) {
-  let shadows = Array.from(splitAtTopLevelOnly(input, ','))
+  let shadows = splitAtTopLevelOnly(input, ',')
   return shadows.map((shadow) => {
     let value = shadow.trim()
     let result = { raw: value }

src/util/splitAtTopLevelOnly.js

@@ -1,5 +1,3 @@
-import * as regex from '../lib/regex'
-
 /**
  * This splits a string on a top-level character.
  *
@@ -15,57 +13,33 @@ import * as regex from '../lib/regex'
  * @param {string} input
  * @param {string} separator
  */
-export function* splitAtTopLevelOnly(input, separator) {
-  let SPECIALS = new RegExp(`[(){}\\[\\]${regex.escape(separator)}]`, 'g')
-
-  let depth = 0
-  let lastIndex = 0
-  let found = false
-  let separatorIndex = 0
-  let separatorStart = 0
-  let separatorLength = separator.length
-
-  // Find all paren-like things & character
-  // And only split on commas if they're top-level
-  for (let match of input.matchAll(SPECIALS)) {
-    let matchesSeparator = match[0] === separator[separatorIndex]
-    let atEndOfSeparator = separatorIndex === separatorLength - 1
-    let matchesFullSeparator = matchesSeparator && atEndOfSeparator
-
-    if (match[0] === '(') depth++
-    if (match[0] === ')') depth--
-    if (match[0] === '[') depth++
-    if (match[0] === ']') depth--
-    if (match[0] === '{') depth++
-    if (match[0] === '}') depth--
-
-    if (matchesSeparator && depth === 0) {
-      if (separatorStart === 0) {
-        separatorStart = match.index
+export function splitAtTopLevelOnly(input, separator) {
+  let stack = []
+  let parts = []
+  let lastPos = 0
+
+  for (let idx = 0; idx < input.length; idx++) {
+    let char = input[idx]
+
+    if (stack.length === 0 && char === separator[0]) {
+      if (separator.length === 1 || input.slice(idx, idx + separator.length) === separator) {
+        parts.push(input.slice(lastPos, idx))
+        lastPos = idx + separator.length
       }
-
-      separatorIndex++
     }
 
-    if (matchesFullSeparator && depth === 0) {
-      found = true
-
-      yield input.substring(lastIndex, separatorStart)
-      lastIndex = separatorStart + separatorLength
-    }
-
-    if (separatorIndex === separatorLength) {
-      separatorIndex = 0
-      separatorStart = 0
+    if (char === '(' || char === '[' || char === '{') {
+      stack.push(char)
+    } else if (
+      (char === ')' && stack[stack.length - 1] === '(') ||
+      (char === ']' && stack[stack.length - 1] === '[') ||
+      (char === '}' && stack[stack.length - 1] === '{')
+    ) {
+      stack.pop()
     }
   }
 
-  // Provide the last segment of the string if available
-  // Otherwise the whole string since no `char`s were found
-  // This mirrors the behavior of string.split()
-  if (found) {
-    yield input.substring(lastIndex)
-  } else {
-    yield input
-  }
+  parts.push(input.slice(lastPos))
+
+  return parts
 }

tests/arbitrary-values.test.css

@@ -661,6 +661,9 @@
 .bg-opacity-\[var\(--value\)\] {
   --tw-bg-opacity: var(--value);
 }
+.bg-\[linear-gradient\(to_left\2c rgb\(var\(--green\)\)\2c blue\)\] {
+  background-image: linear-gradient(to left, rgb(var(--green)), blue);
+}
 .bg-\[url\(\'\/path-to-image\.png\'\)\] {
   background-image: url('/path-to-image.png');
 }

tests/arbitrary-values.test.html

@@ -220,6 +220,7 @@
     <div class="bg-[#0f0_var(--value)]"></div>
     <div class="bg-[var(--value1)_var(--value2)]"></div>
     <div class="bg-[color:var(--value1)_var(--value2)]"></div>
+    <div class="bg-[linear-gradient(to_left,rgb(var(--green)),blue)]"></div>
 
     <div class="bg-[url('/path-to-image.png')] bg-[url:var(--url)]"></div>
     <div class="bg-[linear-gradient(#eee,#fff)]"></div>