The main advantage of the logging mechanism is that it easily hooks into -v / --very-verbose to enable higher levels of logging. Once we have that, I think it makes sense to use the logging infrastructure for all messages just for code consistency, just have fatal errors go to a level that we always print out.

So I think we should go the following route in the tools:

• never use print
• never use stderr.writeln/stdout.writeln
• do all logging using the logging infrastructure
• have four levels of logging:
  • fatal errors, which are always output on stderr
  • normal messages (primarily warnings or high-important priority messages like "No analyzer warnings"), which are always output on stdout (maybe we can have a -q option to quiet them)
  • information that's output to stdout if you enable -v mode
  • verbose information that's output to stdout if you enable --very-verbose mode