[akhilcb]

Motivation and Context

Data.parseDataURL decided whether a data URL is base64 with
string.contains(";base64,"), which searches the payload too, not just the header. A
plain-text data URL whose content contains ;base64, was treated as base64, failed to
decode, and returned nil:

data:text/plain,hello;base64,world  →  nil   (expected: "hello;base64,world")

isDataURL returns true for the same string, so the two functions disagreed.

The data URL regex now captures the optional ;base64 marker, and that capture decides
whether to decode base64. The payload is never searched.

How Has This Been Tested?

Adds tests for:

• the input above, also checking that isDataURL and parseDataURL agree;
• a regression test confirming that standard ;base64, data URLs still decode as base64;
• ;base64 following a charset parameter (data:text/plain;charset=utf-8;base64,SGk=).

swift test passes.

Breaking Changes

None. Value decoding follows parseDataURL, so a JSON string like the one above now
decodes as .data, as other data-URL strings already do.

Types of changes

• Bug fix (non-breaking change which fixes an issue)

[varun06]

Thanks @akhilcb, the fix looks right to me. Capturing the ;base64 marker, which the regex already
limits to the header, is exactly what's needed, and the new test fails on main without the fix.

One request: could you drop the last commit (test: disable reconnection in the heartbeat failure test)?

• It's unrelated to data URLs.
• With reconnection disabled, nothing in the transport touches the connection after the simulated failure, so the test only checks the state the mock set itself.
• It also conflicts with #283: once that cancels the connection when reconnection is disabled, the state becomes .cancelled and this test fails. I reproduced that locally.

If the test is flaky as written, a separate PR that asserts the transport's behavior would be great.

Minor: once #295 lands, Value no longer turns data URLs into .data, so the "Breaking Changes" note can be updated then.

[akhilcb]

@varun06 Thanks for the review.

I have dropped the heartbeat commit, so this PR now has only the two data URL commits. I also rebased it onto the latest main while I was at it.

You are right that the reconnection change doesn't belong here and good catch on #283. With reconnection disabled, the test only checks the state the mock set itself, so it isn't testing anything in the transport. I'll take another look and open a separate PR that asserts the transport's behavior instead. I will update the Breaking Changes note once #295 lands.

[akhilcb]

I have raised this PR based on our above discussion: #311. Thank you!