Gilles commented on LANG-1373:

{quote}I don't think throwing an exception or ? to get the Stack info just to 
keep from naming a timing is in line with the purpose of the feature, for 
performance reasons.
No need to throw the exception, just instantiate it and get the trace. True, it 
has a cost.
 But the benefit in ease of use is perhaps worth it; moreover, the cost will 
probably decrease with Java 9 ({{StackWalker}}).
 What are the performance requirements?
{quote}The user doesn't have to figure out a naming scheme. They just have to 
give the timing a name.
That's what I meant by "naming scheme"; the use-cases shown in the unit tests 
seem to confirm that such a scheme, although required by the current API, is 
actually not necessary.
void functionOneOne(StackWatch<String> watch)  {
    watch.startTiming("OneOne", "OneFunc");
    // ...
void functionOneTwo(StackWatch<String> watch)  {
    watch.startTiming("OneTwo", "OneFunc");
    // ...
In the above excerpt, the "name" looks redundant:
 * function name = "functionOneOne" -> timing name = "OneOne"
 * function name = "functionOneTwo" -> timing name = "OneTwo"

Auto-generating the "name" simplifies the API and makes the usage more robust, 
in a sense, since the user cannot make the mistake of naming the timings in an 
inconsistent way.
 From the following sequence of statements:
public void testStackWatchCallStackNames() {
    StackWatch<String> watch = new StackWatch<>("testStackWatch");
(no "name" argument to the {{startTiming}} methods), the user would get (from 
visiting all the nodes):

> Stopwatch based capability for nested, named, timings in a call stack
> ---------------------------------------------------------------------
>                 Key: LANG-1373
>                 URL: https://issues.apache.org/jira/browse/LANG-1373
>             Project: Commons Lang
>          Issue Type: New Feature
>          Components: lang.time.*
>            Reporter: Otto Fowler
>            Assignee: Otto Fowler
>            Priority: Major
> While working on adding some timing functionality to a Metron feature, I came 
> across the
> Stopwatch class, but found that it didn’t suite my needs.
> What I wanted to do was to create a timing from a top level function in our 
> Stellar dsl, and have have a group of related timings, such that the end 
> result was the overall time of the call, and nested timings of other calls 
> executed during the dsl execution of that function. These timings would all 
> be named, and have a path for identification and include timing the language 
> compiler/execution as well as the function execution itself. It would be 
> helpful if they were tagged in some way as well, such that the consumer could 
> filter during visitation.
> So I have written StackWatch to provide this functionality, and submitted it 
> in a Metron PR.
> From the PR description:
> StackWatch
> A set of utility classes under the new package stellar.common.timing have 
> been added. These provide the StackWatch functionality.
> StackWatch provides an abstraction over the Apache Commons StopWatch class 
> that allows callers to create multiple named and possibly nested timing 
> operations.
> <…>
> This class may be more generally useful to this and other projects, but I am 
> not sure where it would live since we wouldn’t want it in common.
> StackWatch uses a combination of Deque and a custom Tree implementation to 
> create, start and end timing operations.
> A Visitor pattern is also implemented to allow for retrieving the results 
> after the completion of the operation.

This message was sent by Atlassian JIRA

Reply via email to