RE: proposal for ICU tracing
"Carl W. Brown" <[email protected]>
| Newsgroups | gmane.comp.lib.icu.general |
|---|---|
| Message-ID | <[email protected]> |
Tex, > There is a speed advantage to the wrapper approach if it can be > turned on for > some callers and not others, compared to the "on for all/off for > all approach" > of putting it in the lower layers. > The speed cost is more than the if statement, as it includes the > logging of > calls the user is not interested in- so there is the cost of > logging and the > cost of filtering or sorting thru the uninteresting data. This is the point of putting the trace controls in TLS. Usually in most cases it is only specific events that you want to trace. Selectively tracing when the thread is making certain call or under special circumstances will dramatically reduce the overhead without having to reengineer every API to pass trace controls which would be a significant rewrite of ICU. Carl