Cheaper Dependencies#log_call to resolve irritating error message w/ rake db:create

I've just had the irritating experience of:

$rake db:create
rake aborted!
Unknown database 'foo'

... which resulted from a plugin's ActiveRecord model referencing an unknown constant like so:

class Tag < ActiveRecord::Base
  class Jail < Safemode::Jail
    allow :id, :name
  end
end

At this point the class Safemode::Jail was not defined yet, so Dependencies#load_missing_constant kicked in.

Dependencies#load_missing_constant first calls #log_call which builds a log message and sends it to #log. While building the log message it calls #inspect on the ActiveRecord model which tries to look up the table and then fails with the message "Unknown database 'foo'".

The error message above was obviously a little confusing :)

But I also wondered why that whole log message is composed for every call to several Dependencies methods even though #log would only use it (and actually log something) if RAILS_DEFAULT_LOGGER is defined and the log_activity flag is set to true (defaults to false).

  def log_call(*args)
    arg_str = args.collect(&:inspect) * ', '
    /in `([a-z_\?\!]+)'/ =~ caller(1).first
    selector = $1 || '<unknown>'
    log "called #{selector}(#{arg_str})"
  end

  def log(msg)
    if defined?(RAILS_DEFAULT_LOGGER) && RAILS_DEFAULT_LOGGER && log_activity
      RAILS_DEFAULT_LOGGER.debug "Dependencies: #{msg}"
    end
  end

I suggest to skip the log message building part in #log_call for those cases when #log wouldn't do anything anyways:

  def log_call(*args)
    return unless defined?(RAILS_DEFAULT_LOGGER) && RAILS_DEFAULT_LOGGER && log_activity
    ...

This would also allow Dependencies to actually load the missing constant in that scenario mentioned above.

Additionally I wonder if it's a good idea to change ActiveRecord::Base#inspect so that it doesn't bomb when the database does not exist?