I'm going to need it to enable multiple-globals-per-compartment in Gecko.

I looked at this briefly today, and there are some complications:

1. newGlobal() expects to synchronously return a global. The only way to synchronously create a global on the web is to create an initial about:blank, but that always inherits the origin of the creator, and hence in our compartment-per-origin world will be same-compartment.

2. The requirement is that newGlobal() return a global that the caller can touch. Which means that in terms of web technologies, we'd need to start with a different-origin thing and then make it accessible (e.g. via document.domain). And the load would have to be async.

So sticking to web-exposed stuff and making this work doesn't really seem possible. Some options:

• Add some sort of privileged-only way to tell an iframe to not share compartments with its parent (e.g. via attribute). It's not clear to me whether this harness runs with any interesting privileges that we could gate that one, and it would require special Gecko code to support this weird mode.

• Just skip the tests that need newCompartment when running jsreftest.

Any other bright ideas?

(In reply to Boris Zbarsky [:bzbarsky, bz on IRC] from comment #0)

• Just skip the tests that need newCompartment when running jsreftest.

That makes sense for now I think, it's just one test? We could mark that one shell-only. Eventually maybe the proposed Realms API will help here or we could do something more involved.

Only these two TypedArray tests need to be modified. And because I've originally written them, I think it's only fair that I also update them to work with same-compartment globals. :-)

Drive-by fixes:

• Remove dynamic check if newGlobal is present, because it is no longer needed (no longer shell-only) and other tests already require it to be present.
• Replace dynamic check for Debugger with |skip-if| guard to ensure we're running in the shell.

Updated patch to remove global. in typeof global.window test.

Hmm, I probably should have written a patch which works with and without multiple globals per compartment. I'll update it later and request a new review.

Alternative patch which works in the browser runner with and without multiple globals per compartment. This allows to check in the patch in current central without waiting for the multiple globals per compartment work.

There is already another test which accepts a TypeError from either global, so let's do the same here, too.

I've already checked the new patch works with current central and with bz's patch queue for the multiple globals per compartment work.

Pushed by ncsoregi@mozilla.com:
https://hg.mozilla.org/integration/mozilla-inbound/rev/91023d410af1
Add fallback because browser "newGlobal()" doesn't support the "newCompartment" option. r=jandem

https://hg.mozilla.org/mozilla-central/rev/91023d410af1