Hello NG,

I'm using Hyades (Release Version 3.2.0) to profile some combinatorial

optimization algorithm. When examining the profiling results, some

oddities catched my eye.

1) In several classes there is a discrepancy between the time spent in

a class and the time spent in methods. Example: A class has only one

method which does some calculations. The package-centric statistics

view states 7,484 for the method (base time) and 0,003 for the

constructor. The base time for the class itself is 7,086. Shouldn't the

time for the class be the sum of the time for the single method and the

time for the constructor? The same discrepancy appears for the

cumulative time.

2) The second oddity is related to the same view. The view states the

time spent in some particular packages. The problem I see is, that the

cumulative time is calculated by summing the cumulative time for all

classes in a package. In my eyes this is simply wrong and produces

fantasy values. An example:

package/class/method cumulative time

-------------------- ---------------

Package A 250

Class A 150

calc(), calls B.calc() 150

Class B 100

calc() 100

Package B

Class C

main(), calls A.calc() 150

C ist the main method of the application and calls A.calc(). Inside

A.calc() the calc()-method of B is called which does all the work.

Overall the whole application spents 150 on its execution, but the

statistics view says 250 is spent in package A (longer than the whole

application!), because it sums the values of A and B.

I hope, someone is able to clarify these things for me.

Best regards,

Andreas Rummler]]>

Thanks for reporting this problem, I opened Bugzilla report:

https://bugs.eclipse.org/bugs/show_bug.cgi?id=91124 and I will follow up

there, I added yourself to the CC list.

Thanks,

Marius

"Andreas Rummler" <andreas.rummler@tu-ilmenau.de> wrote in message

news:dbcc2df8436c5c18994e6c14cab8e440$1@www.eclipse.org...

> Hello NG,

>

> I'm using Hyades (Release Version 3.2.0) to profile some combinatorial

> optimization algorithm. When examining the profiling results, some

> oddities catched my eye.

>

> 1) In several classes there is a discrepancy between the time spent in

> a class and the time spent in methods. Example: A class has only one

> method which does some calculations. The package-centric statistics

> view states 7,484 for the method (base time) and 0,003 for the

> constructor. The base time for the class itself is 7,086. Shouldn't the

> time for the class be the sum of the time for the single method and the

> time for the constructor? The same discrepancy appears for the

> cumulative time.

>

> 2) The second oddity is related to the same view. The view states the

> time spent in some particular packages. The problem I see is, that the

> cumulative time is calculated by summing the cumulative time for all

> classes in a package. In my eyes this is simply wrong and produces

> fantasy values. An example:

>

> package/class/method cumulative time

> -------------------- ---------------

>

> Package A 250

> Class A 150

> calc(), calls B.calc() 150

> Class B 100

> calc() 100

>

> Package B

> Class C

> main(), calls A.calc() 150

>

>

> C ist the main method of the application and calls A.calc(). Inside

> A.calc() the calc()-method of B is called which does all the work.

> Overall the whole application spents 150 on its execution, but the

> statistics view says 250 is spent in package A (longer than the whole

> application!), because it sums the values of A and B.

>

> I hope, someone is able to clarify these things for me.

>

> Best regards,

>

> Andreas Rummler

>]]>