i don't like that patch for two reasons: - if config_pending must be exposed, do it properly, not with a another extern that isn't seen by the definition. subr_prf.c is already gross enough in that way. however, i think it is an abuse of it to use it this way, early autoconfig can happen without ever increasing config_pending -- it all depends upon what devices you have configured and what are present.
- no prefix at all seems worse. at least i know it was an error before, but no there is no context, just a message. however, the biggest problem, IMO, is the presence of the API aprint_error() -- it takes no "device" parameter for a name, and thus is prefixless in messages making them confusing. can we eliminate this one entirely while we are at it? it's unfortunately used a *lot*. > 2. We don't have a non-autoconfig-related family of printf > calls to handle errors outside autoconfiguration. we have device_printf(9). perhaps a device_printf_error()? .mrg.
