tailwind-merge is built to be intuitive. It follows a set of rules to determine which class wins when there are conflicts. Here is a brief overview of its conflict resolution.
twMerge('p-5 p-2 p-4') // → 'p-4'
twMerge('p-3 px-5') // → 'p-3 px-5'
twMerge('inset-x-4 right-4') // → 'inset-x-4 right-4'
twMerge('inset-x-px -inset-1') // → '-inset-1'
twMerge('bottom-auto inset-y-6') // → 'inset-y-6'
twMerge('inline block') // → 'block'
twMerge('p-2 hover:p-4') // → 'p-2 hover:p-4'
twMerge('hover:p-2 hover:p-4') // → 'hover:p-4'
twMerge('hover:focus:p-2 focus:hover:p-4') // → 'focus:hover:p-4'
The order of standard modifiers does not matter for tailwind-merge.
twMerge('bg-black bg-[color:var(--mystery-var)]') // → 'bg-[color:var(--mystery-var)]'
twMerge('grid-cols-[1fr,auto] grid-cols-2') // → 'grid-cols-2'
Warning Labels necessary in ambiguous cases
When using arbitrary values in ambiguous classes like
text-[calc(var(--rebecca)-1rem)]
tailwind-merge looks at the arbitrary value for clues to determine what type of class it is. In this case, like in most ambiguous classes, it would try to figure out whethercalc(var(--rebecca)-1rem)
is a length (making it a font-size class) or a color (making it a text-color class). For lengths it takes clues into account like the presence of thecalc()
function or a digit followed by a length unit like1rem
.But it isn't always possible to figure out the type by looking at the arbitrary value. E.g. in the class
text-[theme(myCustomScale.rebecca)]
tailwind-merge can't know the type of the arbitrary value and will default to a text-color class. To make tailwind-merge understand the correct type of the arbitrary value in those cases, you can use CSS data type labels which are used by Tailwind CSS to disambiguate classes:text-[length:theme(myCustomScale.rebecca)]
.
twMerge('[mask-type:luminance] [mask-type:alpha]') // → '[mask-type:alpha]'
twMerge('[--scroll-offset:56px] lg:[--scroll-offset:44px]')
// → '[--scroll-offset:56px] lg:[--scroll-offset:44px]'
// Don't do this!
twMerge('[padding:1rem] p-8') // → '[padding:1rem] p-8'
Warning tailwind-merge does not resolve conflicts between arbitrary properties and their matching Tailwind classes to keep the bundle size small.
twMerge('[&:nth-child(3)]:py-0 [&:nth-child(3)]:py-4') // → '[&:nth-child(3)]:py-4'
twMerge('dark:hover:[&:nth-child(3)]:py-0 hover:dark:[&:nth-child(3)]:py-4')
// → 'hover:dark:[&:nth-child(3)]:py-4'
// Don't do this!
twMerge('[&:focus]:ring focus:ring-4') // → '[&:focus]:ring focus:ring-4'
Warning Similarly to arbitrary properties, tailwind-merge does not resolve conflicts between arbitrary variants and their matching predefined modifiers for bundle size reasons.
The order of standard modifiers before and after an arbitrary variant in isolation (all modifiers before are one group, all modifiers after are another group) does not matter for tailwind-merge. However, it does matter whether a standard modifier is before or after an arbitrary variant both for Tailwind CSS and tailwind-merge because the resulting CSS selectors are different.
twMerge('!p-3 !p-4 p-5') // → '!p-4 p-5'
twMerge('!right-2 !-inset-x-1') // → '!-inset-x-1'
twMerge('text-sm leading-6 text-lg/7') // → 'text-lg/7'
twMerge('p-5 p-2 my-non-tailwind-class p-4') // → 'my-non-tailwind-class p-4'
twMerge('text-red text-secret-sauce') // → 'text-secret-sauce'
tailwind-merge has some features that simplify composing class strings together. Those allow you to compose classes like in clsx, classnames or classix.
twMerge('some-class', 'another-class yet-another-class', 'so-many-classes')
// → 'some-class another-class yet-another-class so-many-classes'
twMerge('some-class', undefined, null, false, 0) // → 'some-class'
twMerge('my-class', false && 'not-this', null && 'also-not-this', true && 'but-this')
// → 'my-class but-this'
twMerge('some-class', [undefined, ['another-class', false]], ['third-class'])
// → 'some-class another-class third-class'
twMerge('hi', true && ['hello', ['hey', false]], false && ['bye'])
// → 'hi hello hey'
Why no object support? Read here.
tailwind-merge is optimized for speed when running in the browser. This includes the speed of loading the code and the speed of running the code.
Results get cached by default, so you don't need to worry about wasteful re-renders. The library uses a computationally lightweight LRU cache which stores up to 500 different results by default. The cache is applied after all arguments are joined together to a single string. This means that if you call twMerge
repeatedly with different arguments that result in the same string when joined, the cache will be hit.
The cache size can be modified or opt-out of by using extendTailwindMerge
.
Expensive computations happen upfront so that twMerge
calls without a cache hit stay fast.
The initial computations are called lazily on the first call to twMerge
to prevent it from impacting app startup performance if it isn't used initially.
Next: Limitations
Previous: What is it for