On Thu, 2007-05-10 at 00:38 +0100, John O'Hara wrote:
> Fancy backends?
>
> Its "only" logging - I've found the entire world resolves down to files and
> syslog; any fanciness tends to go the way of the dinosaur by being too
> complex for joe average </speaking-from-bitter-experience>
>
> With files (pipes) and syslog its amazing what you can do.
>
> I do like the rlog idea -- seems very "dtrace" in its conception. Cool, but
> too cool for its own good?
> As you say, thread nastiness may await where a hardware platform cant
> replace a function pointer in one atomic memory write.
Yup, I've decided going for simplicity and replacability. Simple
one-macro QPID_LOG API with two initial implementations, rlog and
braindead global trace level int to keep Apache happy (rlog is LGPL.) We
can replace/extend the set of logging options any time.
Both of these provide the minimal if (int/pointer-compare) {} overhead
for disabled log statements. The pantheios API is dead clever but it's
overhead for unlogged calls is no better, it has a bizzaro build system
and hasn't been packaged in a form we could depend on so would be work
to integrate. Its formatting performance for logged calls is better but
right now I want to get something working ASAP that won't kill
performance with logging off and we can improve later if need be.
