Alistair Francis <alistair.fran...@xilinx.com> writes: > On Wed, Jun 7, 2017 at 12:19 AM, Markus Armbruster <arm...@redhat.com> wrote: >> Paolo Bonzini <pbonz...@redhat.com> writes: >> >>> On 06/06/2017 18:30, Alistair Francis wrote: >>>>> >>>>> This is somehow confusing. I don't think it is worth having another >>>>> qemu_log_stderr() function rather than using error_report() but this very >>>>> call might deserve a comment explaining this unusual use. What do you >>>>> think? >>>> >>>> The problem with stderr is that this isn't an error. Some uses of QEMU >>>> (inside Eclipse for example) flag everything printed on stderr as red >>>> which confuses users that they are seeing an error when they really >>>> aren't. >>> >>> But they are wrong. >> >> Concur. We also print warnings and informational messages to stderr. >> >> We should make errors easy to recognize. Fortunately, error_report() >> prints errors to stderr in a rigid format. Unfortunately, error >> messages bypassing error_report() still exist in places. We suck. >> >> The format is >> >> timestamp-if-enabled progname ':' location message >> >> timestamp-if-enabled is normally empty. With -msg timestamp=on, it's >> the current time in ISO 8601 format, followed by a space. >> >> progname is the program name (main()'s argv[0]). >> >> location is either empty, or a reference to the command line or a >> configuration file. >> >> See error_vreport() for details. > > Ok, but this isn't an error, it's more information. So it sounds like > we should still print to stderr but not print in the format described > above?
Yes. I explained the error message format to show how to distinguish actual errors from other stuff.
