There are currently two command line options that can be used:

 --remote-debugger [<host>][:<port>]
 --remote-debugging-port <port> Start the Firefox remote agent, which is
                     a low-level debugging interface based on the CDP protocol.
                     Defaults to listen on localhost:9222.

Not sure why we actually went with both (especially the first is confusing regarding devtools), and not just used the --remote-debugging-port one to be in sync with Chrome, which has support for the following arguments:

--remote-debugging-address ⊗ 	Use the given address instead of the default loopback for accepting remote debugging connections. Should be used together with --remote-debugging-port. Note that the remote debugging protocol does not perform any authentication, so exposing it too widely can be a security risk. ↪
--remote-debugging-pipe ⊗ 	Enables remote debug over stdio pipes [in=3, out=4]. Optionally, specifies the format for the protocol messages, can be either "JSON" (the default) or "CBOR". ↪
--remote-debugging-port ⊗ 	Enables remote debug over HTTP on the specified port. ↪
--remote-debugging-socket-name[7] ⊗ 	Enables remote debug over HTTP on the specified socket name. ↪

I would suggest that we rename --remote-debugger to --remote-debugging-address.

This change shouldn't cause trouble given that --remote-debugger isn't used by Puppeteer.

Maja and James, what's your take?

Actually if we do how would we determine if the remote agent needs to be started? Maybe we have to rename --remote-debugger to just --remote-debugging? I wonder how chromedriver enables it.

If I'm not wrong they only set the --remote-debugging-port property:
https://source.chromium.org/chromium/chromium/src/+/master:chrome/test/chromedriver/test/run_py_tests.py;l=4291?q=remote-debugging-port&ss=chromium%2Fchromium%2Fsrc:chrome%2Ftest%2Fchromedriver%2F

It seems reasonable to me to make it just --remote-debugging-port; I don't think we should support non-localhost hostnames at this point.

In an ideal world we'd use a totally different mechaism that allows firefox to pick an arbitary port and return that to the client somehow. But that's presumably out of scope for this bug given that we don't want to break puppeteer.

If nothing depends on --remote-debugger we could just remove it.

I will have a look at it tomorrow.

Using different names for enabling the remote agent via
--remote-debugger and --remote-debugging-port adds confusion
to users. Also given that --remote-debugging is actually
not used by any client library including Puppeteer,
it can be removed.

Using --remote-debugging-port instead will bring us onpar
with Chrome and how it enables CDP.

Henrik, could you fix the examples in the docs to reflect that a port number is required? e.g. --remote-debugging-port=0

Huh, nevermind. Locally, my artifact build from autoland doesn't reflect this change yet.

https://hg.mozilla.org/mozilla-central/rev/1616add3a380