fix(css-vars): nullish v-bind in style should not lead to unexpected inheritance

[Justineo]

• Fixes #12434
• Fixes #12439
• Fixes #7474
• Close #7475

Currently, we generate CSS custom properties on component roots for each v-bind in SFC <style> blocks. When one component instance is nested inside another, the nested instance can inherit values from the outer instance, even if its own binding value is undefined or null. Ideally, style bindings should be scoped per component instance since they depend on the instance's state. However, with the current approach using CSS custom properties, we can't generate unique hashed names for each instance because CSS can only reference a single property.

To address this, the improvement here is to "reset" these custom properties when the binding value is nullish.

Current Behavior:

• In SSR, we ignore nullish values.
• In CSR, we generate --<hash>-<name>:undefined for undefined but ignore null, which causes inconsistencies and is related to the hydration warning in #12439.

This PR:

• Ensures both CSR and SSR generate --<hash>-<name>:initial for nullish values. This properly resets the custom property, as explained in the CSS spec. It behaves as if no custom property is defined in ancestor elements.

SSR Note:

• Added a special : prefix to distinguish style entries coming from v-bind in styles. This allows us to process them separately from user-defined custom properties in ssrRenderStyle. Alternatively, we can also create a new runtime helper and modify the codegen here:

Not sure if my current approach is preferred. Let me know if you have other opinions.

[github-actions[bot]]

Size Report

Bundles

File	Size	Gzip	Brotli
runtime-dom.global.prod.js	100 kB (+42 B)	38 kB (+21 B)	34.2 kB (+16 B)
vue.global.prod.js	158 kB (+42 B)	57.8 kB (+22 B)	51.5 kB (+30 B)

Usages

Name	Size	Gzip	Brotli
createApp (CAPI only)	47.2 kB	18.3 kB	16.8 kB
createApp	55.2 kB	21.3 kB	19.5 kB
createSSRApp	59.3 kB	23.1 kB	21 kB
defineCustomElement	60.1 kB	22.9 kB	20.8 kB
overall	69.1 kB	26.5 kB	24.1 kB

[pkg-pr-new[bot]]

Open in Stackblitz

@vue/compiler-core

``` pnpm add https://pkg.pr.new/@vue/compiler-core@12461 ```

@vue/compiler-dom

``` pnpm add https://pkg.pr.new/@vue/compiler-dom@12461 ```

@vue/compiler-ssr

``` pnpm add https://pkg.pr.new/@vue/compiler-ssr@12461 ```

@vue/compiler-sfc

``` pnpm add https://pkg.pr.new/@vue/compiler-sfc@12461 ```

@vue/reactivity

``` pnpm add https://pkg.pr.new/@vue/reactivity@12461 ```

@vue/runtime-core

``` pnpm add https://pkg.pr.new/@vue/runtime-core@12461 ```

@vue/runtime-dom

``` pnpm add https://pkg.pr.new/@vue/runtime-dom@12461 ```

@vue/server-renderer

``` pnpm add https://pkg.pr.new/@vue/server-renderer@12461 ```

@vue/shared

``` pnpm add https://pkg.pr.new/@vue/shared@12461 ```

vue

``` pnpm add https://pkg.pr.new/vue@12461 ```

@vue/compat

``` pnpm add https://pkg.pr.new/@vue/compat@12461 ```

commit: 4b8e266

[Justineo]

/ecosystem-ci run

[vue-bot]

📝 Ran ecosystem CI: Open

suite	result	latest scheduled
language-tools	:white_check_mark: success	:white_check_mark: success
nuxt	:white_check_mark: success	:white_check_mark: success
pinia	:white_check_mark: success	:white_check_mark: success
primevue	:white_check_mark: success	:white_check_mark: success
quasar	:white_check_mark: success	:white_check_mark: success
radix-vue	:white_check_mark: success	:white_check_mark: success
router	:white_check_mark: success	:white_check_mark: success
test-utils	:white_check_mark: success	:white_check_mark: success
vant	:white_check_mark: success	:white_check_mark: success
vite-plugin-vue	:white_check_mark: success	:white_check_mark: success
vitepress	:white_check_mark: success	:white_check_mark: success
vue-i18n	:white_check_mark: success	:white_check_mark: success
vue-macros	:white_check_mark: success	:white_check_mark: success
vuetify	:white_check_mark: success	:white_check_mark: success
vueuse	:x: failure	:x: failure
vue-simple-compiler	:white_check_mark: success	:white_check_mark: success

[skirtles-code]

I think this would also close #7474 and the associated PR #7475.

---

There's an underlying question here about whether we consider undefined and null to be valid values for CSS v-bind(). Should we support them, or should we just log a warning? It is possible to handle them in userland code instead, e.g. using v-bind(c ?? 'initial').

Then there's a follow up decision about where to draw the line. What about an empty string (Playground)? Or a white-space value like " "? Or an empty array (Playground)? Etc.

I don't want to expand the scope of this PR unnecessarily, but I do think it's important to be clear why we're drawing the line at nullish values.

[Justineo]

@skirtles-code Thanks for the feedback! Great questions!

Here’s what I thought:

1. Why not handle them in userland? Theoretically, using CSS custom properties for v-bind in <style> blocks is an implementation detail that users are not supposed to be aware of. However, there are notable differences in behavior between color: v-bind('"initial"') and color: initial. The former binds initial to the custom property, making it functionally equivalent to color: unset, whereas the latter directly applies the initial keyword as intended in CSS. Making users to write initial to reset values while it actually behaves like unset can be quite frustrating, so I would prefer to make it more transparent.
2. Why only convert nullish values? You're absolutely right—more values should be considered. I think it's kind of similar to our props coercion, where we are removing a binding with nullish values. For CSS custom properties, "removing" in this context means resetting to their initial value, which is the guaranteed invalid value, which can be explicitly set with the keyword initial. Per the CSS specification, empty strings and white spaces for custom properties behave differently from initial, as they prevent fallback values from taking effect. Currently, we are removing the custom property for empty strings, but retaining white spaces. This creates an inconsistency if users are binding the value to a custom property themselves (Playground). To address this, I believe we should treat empty strings and white spaces as-is, maintaining consistency and aligning better with user expectations. I reconsidered and think we should handle it comprehensively as outlined in the table below. Could you kindly help review it? Thank you!

The situation is more complicated than I expected. I'm investigating the current behavior of Vue and CSS custom properties.

Current behavior:

v-bind value	SSR output	CSR output	Mismatch
null	/	/
undefined	/	--foo: undefined	⚠️
empty string	--foo:;	/	⚠️
white spaces	--foo: ;	--foo: ;

For Vue, undefined and empty string will cause SSR/CSR mismatch

For CSS custom properties:

• The cssText, code in <style>, and style attribute behave differently from style.setProperty. During SSR, we use the former, while during CSR, we use both.
• The behavior of --foo:; differs from style.setProperty('--foo', ''). The former acts like white spaces, while the latter effectively removes the declaration.

Since we are binding values to CSS, we should align with CSS behavior, where an empty string is treated as equivalent to white spaces. To fix the SSR/CSR mismatch and ensure consistent resetting, I propose handling values as follows:

v-bind value	SSR	CSR
null	--foo: initial	style.setProperty('--foo', 'initial')
undefined	--foo: initial	style.setProperty('--foo', 'initial')
empty string	--foo: ;	style.setProperty('--foo', ' ')
white spaces	--foo: ;	style.setProperty('--foo', ' ')

For other string values, we should output them as-is. For unsupported types of values, we should warn about the binding and retain the current behavior (rendering everything as a string, likely using String(value)).

WDYT? /cc @adamdehaven @skirtles-code @edison1105

[adamdehaven]

@Justineo I thought we determined that a whitespace character would be ideal?

[Justineo]

After rereading the spec, I believe that only the initial value truly resets the custom property to its default state, as if it were never set.

[adamdehaven]

After rereading the spec, I believe that only the initial value truly resets the custom property to its default state, as if it were never set.

The white space character seems to work in practice if you can try?

[Justineo]

The white space character seems to work in practice if you can try?

White space sets a valid empty value, which will prevent fallback value to take effect. initial resets the custom property to its initial value–the guaranteed invalid value, which is the same as it’s not declared. I have tested and can confirm the implementation is aligned with the spec in all major browsers.

[adamdehaven]

The white space character seems to work in practice if you can try?

White space sets a valid empty value, which will prevent fallback value to take effect. initial resets the custom property to its initial value–the guaranteed invalid value, which is the same as it’s not declared. I have tested and can confirm the implementation is aligned with the spec in all major browsers.

I'm not sure injecting initial in all those places though is desirable. Do you have a reproduction with this behavior?

In my testing, initial and a white space character behave differently.

See the example taken from here

[Justineo]

I'm not sure injecting initial in all those places though is desirable.

It is the standard way to reset a custom property.

Do you have a reproduction with this behavior?

https://codepen.io/Justineo/pen/VYZZPOz

As shown in this pen, initial works the same as not setting the value.

In my testing, initial and a white space character behave differently.

That’s why we should use initial instead of white spaces…

See the example taken from here

We certainly don’t want to invalidate the fallback value red in var(--foo, red) when no valid value is bound to --foo. Therefore, using initial is the only appropriate solution.

[adamdehaven]

https://codepen.io/Justineo/pen/VYZZPOz

The reproduction looks correct; thanks for putting that together

[skirtles-code]

TLDR: I think the latest proposal looks good.

---

I wasn't initially clear about the difference between initial and v-bind('"initial"'), so I put together a Playground to help demonstrate:

• Playground

v-bind('"initial"') does indeed behave like unset, not initial. I concur that this does make handling undefined/null in userland less appealing.

---

Using initial for undefined and null makes sense to me.

I also like the idea of treating an empty string as a space. That prevents the variable being inherited from the parent component and also seems consistent with vanilla CSS. It wasn't immediately clear to me why a space character would be better than initial in that case, so I put together another Playground to show an example where it would actually make a difference:

• Playground

In that example, we would want v-bind('""') to behave like .bar and .baz, not .foo, so a space character seems to give us what we want.

[Justineo]

@skirtles-code Thank you for the demos! I’ll update the code to align with our latest conclusions.