GH-1246. Asynchronously initialize cache before reading II

[kotman12]

This is an alternative, simpler implementation to #1247 with the same motivation, namely to solve #1246

[tisonkun]

Thanks for preparing this patch! I'd review it by the end of next week (sorry this week is a busy week for me >_<)

[tisonkun]

Looks good and very tidy.

Thanks a lot for your contribution!

[tisonkun]

I'm waiting for other committers' review (cc @eolivelli @kezhuw @Randgalt @cammckenzie) but would otherwise merge this patch in the next week if there is no more objection.

[kotman12]

Thanks for taking a look @tisonkun!

[kotman12]

Hey .. I just did a walk-through of the code again and have a question. Is applying the filter:

e -> !e.getKey().isRoot()
                                && e.getKey().parent().equals(client.modelSpec().path())

safe to do from ::list()? I realize I added that behavior by deferring to the newly created internalChildren from within list which applies this filter that wasn't applied before. I am worried this is a regression not caught by the tests.

If it is ok then it is probably ok to remove the now unused methods in `ModeledCachedImpl:

    ZPath basePath() {
        return basePath;
    }

    Map<ZPath, ZNode<T>> currentChildren() {
        return currentChildren(basePath);
    }

Either way I think a change is due ... let me know what you think!

[tisonkun]

@kotman12 Let's removed the unused method in a follow-up PR.

I'm going to merge this PR after CI green.

[tisonkun]

Merged. Thanks for your contribution @kotman12!

Would you provide a follow-up to remove the unused methods mentioned above?

[kotman12]

@tisonkun I have done so here #1257 .. looking at the past behavior of ::list as well as re-reading the API description I think I did inadvertently break the contract with this PR. Hopefully the change to fix it makes sense.