On Fri, May 22, 2015 at 12:22 PM, Bobby Holley <bobbyhol...@gmail.com> wrote:
> On Fri, May 22, 2015 at 12:11 PM, Gregory Szorc <g...@mozilla.com> wrote: > >> Better than a log level is an event type (possibly enumerated). When >> people >> are looking at log output, they want to see specific events. While >> filtering by the logger is a good way to limit/expand logging, I find that >> this approach is often limiting, even when log levels are used. This >> especially holds true when using the most verbose settings and output is >> spammy. You end up writing tons of grep chains to sort through the mess. >> > > The problem is that too much logging can have huge costs - when debugging > media races, I often need to limit the test run to very light logging in > order to avoid changing the timing characteristics too much and preventing > the failure from reproducing. There's also the issue that even just a > couple of media tests run at full logging can create hundreds of megabytes > of output. > Of course. I don't think log levels and event types are mutually exclusive. If I were designing a log system from scratch, I'd probably make event types the primary identifier and have mappings of event type to log level or named filters. This gets a little complicated in C++ because you need to apply the filter at compile time to avoid run-time penalties. Trade-offs are hard :/ _______________________________________________ dev-platform mailing list dev-platform@lists.mozilla.org https://lists.mozilla.org/listinfo/dev-platform
