This hasn't fixed the problem for me unfortunately. Here is my test:

# global.cylc
[platforms]
    [[localhost]]
    [[localhost_test]]
        hosts = <a remote platform>
        install target = <remote platform install target>

The play a workflow e.g:

# flow.cylc
[scheduling]
    [[graph]]
        R1 = """
        hello => goodbye
    """

[runtime]
    [[root]]
        script = "sleep 1"
        platform = localhost_test
    [[hello]]
    [[goodbye]]

hello and goodbye both run locally, not respecting the settings in global.cylc.

As suggested, changing localhost_test to localhxst_test works fine. I'll have a look to see if I can spot the problem.

I can't reproduce this.

There are other uses of this interface with host_check=True so this error might be able to crop up in other places?

I'm not sure I understand the logic of host_check in subshell_eval, seems a little strange:

Return localhost if the host is localhost???

I would have thought the subshell_eval should look something like this:

match = command_pattern.match(command) 
if not match:
    # this isn't a command, no subshell evaluation to perform
    return command

Although note that later on we do:

command = os.path.expandvars(command)

Which might need to be moved further up if we intend to support configuring the platform by an environment variable (e.g. platform = $MY_PLATFORM) which I don't think we do?

Which might need to be moved further up if we intend to support configuring the platform by an environment variable (e.g. platform = $MY_PLATFORM) which I don't think we do?

No.

I'm not sure I understand the logic of host_check in subshell_eval,

This function can be called to find the value of a platform or a host. If it's a host_check indicates we've called it from the compatibility mode part of the code, and that it should retain Cylc 7 behavior - that is to run host check on command, which should, confusingly, be an evaluated host name at this point. Since a platform isn't a host this check makes no sense if the command is getting a value for platform.

I can't say for sure without looking more whether the Cylc 7 behavior is sane. But surely we'd have noticed if unreachable hosts caused jobs to run on local?

I'm not sure I understand the logic of host_check in subshell_eval, seems a little strange:

Only being run if subshell eval is supposed to be returning a host.

As for why it's doing even this, I'm afraid that's going into the depths of time/git.

I have put up a draft PR to improve the readability of the subshell eval code, let me know thoughts: wxtim#53

@MetRonnie & @hjoliver - Poke

I think this is still awaiting merge of @MetRonnie 's side PR to your branch @wxtim ?

Tested as working however. (And I agree the affected code is a bit confusing).

I think this is still awaiting merge of @MetRonnie 's side PR to your branch @wxtim ?

Tested as working however. (And I agree the affected code is a bit confusing).

Yes, but I'm assuming that @MetRonnie has marked it draft for a reason