Re: proposal for ICU tracing

Tex Texin wrote:

> It's not clear to me why tracing for collation is provided but not the other
> api, especially if it is as low cost as indicated.

Tracing is going into collation first because we have collation users 
who require it.  It may be in at least some additional parts of ICU for 
2.8, and, if successful, may grow from there.


> On the other hand, it would seem that a long term plan should be in place as
> well as an assessment of user requirements for tracing, so that any overhead
> can be justified by the need and unnecessary overhead clearly eliminated. 

Tracing is definitely driven primarily by user requirements.

The overhead of supporting it is something that needs careful attention, 
  particularly because some very critical functions will be traced - 
strcoll(), for example.   Tracing by using wrapper functions or macros 
is another possibility, but looks to be more complicated for the user, 
with  little if any speed advantage over the approach taken.  And, all 
tracing code can be compiled out with either approach, if no tracing at 
all is wanted.

   -- Andy Heninger
      [email protected]