Why don't we keep the previous version but do:

[[ -z "${_checksum_md5:-}" && $_name == http* ]] ||		
        _checksum_md5="$(__rvm_curl -s -L $_name.md5)"

I think that's incorrect too, it would run curl if $_name does not start with http, even if the checksum is already set. My first commit e370ad6 I think does the minimal fix.

But it seems highly desirable to avoid curl calls entirely if there are already entries in the config files. Since the code iterates over multiple _name, we need to split the loop, so e.g. config files for the second _name (e.g., truffleruby-19.0.0-linux-amd64.tar.gz) are checked before a URL for the first _name (e.g., a URL). I think this approach somewhat concurs with @mpapis's comment: #4650 (comment), i.e., avoiding needless curl calls.

I also find a pair of if a lot more readable and easier to reason about than multiple &&/|| with side effects.

this will never get to checking URL's if there was no checksum locally, __rvm_checksum_none has to be moved to the end of the function, outside of loops

Bash conditions are hard to read.

__rvm_checksum_none()
{
  [[ -z "${_checksum_md5:-}" && -z "${_checksum_sha512:-}" ]]
}

So that returns "true" if both are unset, i.e, it returns "false" if any is set and therefore || return 0 returns if any is set.
This is also done at the function entry BTW, and already before in the loop.

So that seems incorrect in the case we get one of the checksum, but not the other, right?

I think we should return only if both are set then, i.e., replace that with __rvm_checksum_all && return 0. I'll do that.

@mpapis Should this function return 0 or 1 if we found only one of the checksums?
__rvm_checksum_validate_file seems happy if only one is set.

So __rvm_checksum_any && return 0 would be right (and the original __rvm_checksum_none || return 0 too) but __rvm_checksum_any && return 0 is probably easier to read?

I changed to the equivalent __rvm_checksum_any && return 0.