Adding Microbenchmarks to the JDK forest/trees (JEP-230)

[Chris Hegarty]

Staffan,

Having all the benchmarks located in a single place makes sense to me, but this doesn’t necessarily mean that they need their own repository, in the forest.  

If I can build, run, and test ( usual development cycle ) without any dependency on these benchmarks, or their infrastructure, essential working with a partial forest ( without the ‘benchmark’ repository ), then I can see the possible value in having a separate repository ( so I can skip cloning and updating it ). But, I’m not sure if that is a reasonable justification for a new repository, as it is probably at odds with your goals, or maybe not?

-Chris

On 2 Dec 2014, at 19:53, Staffan Friberg <staffan.friberg at oracle.com> wrote:

> Hi,
> 
> (Adding the jdk9-dev list to increase the visibility of the discussion)
> 
> With the multiple sub-repository commit mechanism improved I believe this might be less of an issue. JPRT can push JDK and HS changes at together and the same functionality should be possible to use for this as well. I wonder if the test issue earlier was that it was a completely separate repository outside of the JDK forest, and less of an issue when being part of the same forest as the JDK source code. Perhaps someone from SQE can chime?
> 
> Otherwise the main reason for having a separate sub-repository on the top level is making it easier to find what benchmarks are available and have a single place to add new once avoid any risk of name duplication. JMH is superb in filtering during execution during runtime so running just a single test or a group of tests is very straight forward and the recommended way, rather than having multiple benchmark JARs. It also makes the build process easier as the building can be done using a single Makefile and a single benchmark JAR (actually two, one for JDK 8 compatible tests and one for JDK 9) that can be picked up by automatic performance testing.
> 
> Cheers,
> Staffan
> 
> On 12/02/2014 06:48 AM, roger riggs wrote:
>> Hi Staffan,
>> 
>> An earlier issue was keeping tests in sync with the code under test, hence
>> the use of test directories within each repository.
>> I think a structure in which the benchmarks for some function and the function
>> itself are in the same repository that is easier to understand and maintain.
>> 
>> $.02, Roger
>> 
>> 
>> On 12/1/2014 7:08 PM, Staffan Friberg wrote:
>>> Hi,
>>> 
>>> Hopefully this is the right list for this discussion.
>>> 
>>> As part of adding Microbenchmarks to the OpenJDK source tree, I'm trying to understand how we best would add the benchmark sources to the existing OpenJDK tree structure.
>>> 
>>> Since the microbenchmark suite will cover all parts of the JDK, covering HotSpot, JDK libraries and Nashorn, it would be preferred to add the microbenchmark directory as a new top level directory. Something similar to the following structure. Having "benchmark" as the top-level directory would allow us to later add different types of benchmarks without colliding with the microbenchmark suite.
>>> 
>>> <openjdk-root>/
>>>   benchmark/microbenchmark/...
>>>   hotspot/...
>>>   jdk/...
>>>   nashorn/...
>>> 
>>> With this as the premise I can see the following 3 options for how this could be added to the source code layout
>>> 
>>> 1. Part of jdk-root repository
>>>     * Only makes sense if we want to move in a direction with fewer
>>>       trees (and eventually a single tree)
>>> 2. Part of another already existing tree
>>>     * Not sure if this is possible without converting and moving the
>>>       directory to a subdirectory of that tree
>>> 3. New tree in the forest/tree structure
>>>     * Most logical option as it follows the current setup and structure
>>> 
>>> 
>>> Anyone have any comments and/or concerns on the suggested directory location and the tree structure in option 3.
>>> 
>>> Would the build-dev team be the right group to later help setup a new tree if decided to be the right way to go?
>>> 
>>> Regards,
>>> Staffan
>>> 
>> 
>