explicitly allow '*' in specs for clique_usage:register/2

[hmmr]

else, dialyzer will complain about specs like this:

clique:register_usage(["riak-admin", "tictacaae", "rebuild-schedule", '*'], ...)

[martinsumner]

Can I just check I understand this correctly.

Is there a need to give consistent usage advice regardless what part of the input is:

i.e.

if someone inputs:

riak admin tictacaae storeheads node or
riak admin tictacaae storeheads partition the storeheads_usage() usage should be returned.

This isn't relying on any clever matching rules in ets - it will match because the code will explicitly lookup '*'.

Dialyzer complains because '*' is an atom(). Though $* could have been used - and there would be no dialyzer issue, or is there something I'm missing? Is there a reason for this to be * as an atom() not * as a string()?

[hmmr]

The choice of '*' as an atom is original authors'. Specs with '*' already exist in riak_core, e.g. https://github.com/OpenRiak/riak_core/blob/openriak-3.2/src/riak_core_cluster_cli.erl#L292, meaning changing it to a '$' (or "*") would snowball into a bunch of new PRs (and affect users of clique outside of riak).

I note that the issue dialyzer is taking with the spec is perfectly valid. Exactly why dialyzer in previous OTP versions didn't flag this case, is an interesting question (it is this issue being discussed here, I think).

if someone inputs:

riak admin tictacaae storeheads node or
riak admin tictacaae storeheads partition the storeheads_usage() usage should be >returned.

Without -n node or -p partition, the command assumes current node and -p all. If, with or without those options, a boolean value is given, it is set for the node/partition(s) specified or inferred. If an argument is present but cannot be converted to a boolean, usage is printed (however, this is a matter to discuss in OpenRiak/riak_kv#31).