fix(astro): Fix isHTMLString check failing in multi-realm environments

[rururux]

fixes: #16003

Changes

When using the Astro Container API, passing child elements to a component inside an MDX file could cause those elements to be incorrectly escaped.
This happened because the value instanceof HTMLString check in the isHTMLString function failed to recognize the HTMLString instance correctly.

astro/packages/astro/src/runtime/server/escape.ts

Lines 50 to 52 in 0d24e3b

	export function isHTMLString(value: any): value is HTMLString {
	return value instanceof HTMLString;
	}

(if(isHTMLString(child)) does not evaluate to true, so the program proceeds to the next if statement)

Following Erika’s suggestion, I have updated the logic to check for the presence of Symbol.for('astro:html-string') instead of using instanceof HTMLString.

Testing

Added a test to verify that child elements passed to a component inside an MDX file are not incorrectly escaped.

Docs

N/A, bug fix

[changeset-bot]

🦋 Changeset detected

Latest commit: e97d963

The changes in this PR will be included in the next version bump.

Not sure what this means? Click here to learn what changesets are.

Click here if you're a maintainer who wants to add another changeset to this PR

[codspeed-hq]

Merging this PR will not alter performance

✅ 18 untouched benchmarks

---

_{Comparing rururux:mdx-escape-issue (e97d963) with main (a7e7567)}

[rururux]

I’ve made a few changes.

• Following Erika’s suggestion, I’ve changed the method to use Symbol.for(...) for the check instead of Symbol.toStringTag.
• As a result, Symbol.toStringTag is no longer used in the HTMLString class, so I’ve removed it.
• I found a check for value instanceof HTMLString in markHTMLString, so I changed it to use isHTMLString instead.
• I’ve simplified the structure of the test fixture. I’ve confirmed that the test fails as it did with the previous test fixture before the change to the isHTMLString check logic.

[Princesseuh]

Did you get to see if adding virtual: to the server-app module name fixes this without requiring changes to HTMLString?

[rururux]

I just tested this, and confirmed that even if I change ASTRO_DEV_SERVER_APP_ID to "virtual:astro:server-app", the tests I added this time will not pass unless I apply this PRs change.
I also confirmed that the error with the <Steps> component we discussed on Discord still occurs without this change.

However, regarding the <Steps> component, HMR stopped working properly after this error and the astro:server-app error were resolved, so the issue hasn’t been fully resolved.

[rururux]

It seems that HMR not working when editing MDX files is by design, my apologies.
So, I think the issue with the <Steps> component can be fixed simply by applying the changes in this PR and updating “virtual:astro:server-app”.

[rururux]

Just to follow up on the HMR mention, by "HMR," I specifically meant changes being applied without a full page reload.
The automatic full reload is still working perfectly fine, so there's no issue with the changes being applied.

[ematipico]

Can you reword the changeset to be more user-centric? https://contribute.docs.astro.build/docs-for-code-changes/changesets/#tips-and-examples

[rururux]

I've updated the changeset. does this look good to you?

[ematipico]

Suggested change

	Fixes HTML content being incorrectly escaped as plain text due to a failed `isHTMLString` check
	Fixes HTML content being incorrectly escaped as plain text when rendering a MDX component using the `AstroContainer` APIs.

cc @rururux for future reference

[rururux]

I appreciate you tidying that up for me. Thanks for the reference!

[rururux]

Thank you! ☺️