Pick better defaults for tracePropagationTargets

[lforst]

The current defaults for tracePropagationTargets (['localhost', /^\/(?!\/)/]) run into a few problems:

• If you do a fetch request to the same origin domain without using a short version (i.e. fetch("https://mysameorigin.com/api/some-route") instead of fetch("/api/some-route")) we will currently not attach tracing headers by default.
• If you use localhost with subdomains (e.g. http://cdn.localhost/resource) we will add tracing headers which may trigger CORS even for localhost.
• Currently, we don't attach tracing headers to requests like fetch("//sameorigin.com/") while it would be valid to do so if sameorigin.com is actually the same origin.

Proposal

• We need to be stricter on the localhost validation, e.g. /^(https?|ws):\/\/localhost//
• We need to add a matcher for when the origin is provided and it is the same origin as the current page, e.g. new RegExp(`^${window.location.origin}\/`)
• (Requirement: No longer support IE11) When a URL starts with a /, e.g. //external-origin.com/, we throw it into the URL constructor as follows and use the resulting origin as the base for matching against items in tracePropagationTargets: new URL(target, window.location.origin). That way, we are matching the actual resolved (including the protocol) against our defined tracePropagationTargets and it will work well with our predefined defaults. Maybe there is a case to be made to generally throw the target in the URL constructor before comparing against anything.

[mydea]

WDYT about doing 1. and 2. in v7? You could consider it breaking, but I'd say it's more of a "fix" and probably what people would expect anyhow..?

[lforst]

@mydea There is definitely an argument to doing it already. I wasn't sure if it is technically breaking though 🤔 Touching tPT is kinda scary because the errors that could result from it are very hight impact.

[lforst]

Especially adding a new matcher is very scary because we will attach headers to more requests.

[Baffour]

@lforst amazing thanks for this! I was about to raise that last point as a bug/feature request, before I found this. Our code has been inconsistently using full URLs in some places, and just the path in others, and it took me a long time to realise this is why distributed tracing wasn't working consistently when tracePropagationTargets was defined