I have raised CLOV-1264, could you describe the problem and attach profiler data in this new issue, please?

Marek,
I'll run a profile session with and without the testsources, both using a Shared recorder and see what the differences are. The problem with creating a ton of *.s files is definitely resolved and technically the coverage doesn't really hang although it's so slow we can't really use it.

As for 3, I'm not sure I have that kind of control over the Grails build process. I'll see what I can do.

Stay tuned. You want to open a separate performance ticket for this issue since this one is closed?

Hi Aaron,

> I'll try profiling a different way to see if I can tell where the time is spent.

1) Could you attach profiler results? Tip: regarding tuning of the profiler - it might be worth to focus on GrowableCoverageRecorder, GrowableCoverageRecorder$FixedProxy, SharedCoverageRecorder, CoverageMatrix, PerTestRecorder$Diffing classes as well as static initialization blocks in other classes.

> With the above configuration it's been running for about 3 hours and is about half finished.

2) Does it mean that the problem is improved, compared to the FIXED recorder? I remember that you wrote that in case of FIXED recorder it took a lot of memory and time during report generation and it was unable to complete the task.

3) I've got one suspicion what could be wrong... Could you try to change your build in such way that the last class instrumented will be loaded into JVM as first one? For example: run the last compiled unit test as first. This might reduce number of memory reallocations due to growing coverage array.

Cheers
Marek

I'm running with the latest snapshot and it doesn't seem to be creating all the .s files, however it's almost unbearably slow unless I disable testsources (as I was doing before). My setuptask is:

    setuptask = { ant, binding, plugin ->
        ant.'clover-setup'(initstring: ".clover/common.db") {
            ant.fileset(dir: "grails-app", includes: "**/domain/**, **/controllers/**, **/jobs/**, **/services/**, **/taglib/**, **/utils/**") { }
            ant.fileset(dir: "src", includes: "**/*.groovy, **/*.java", excludes: "**/script@*.groovy, **/system/generator/**, **/system/database/generators/**, **/castor/**") { }
            ant.fileset(dir: "test", includes: "**/*.groovy") { }
            // ant.testsources(dir: "test", enabled: false) { }
            ant.profiles {
                ant.profile(name: "default", coverageRecorder: "SHARED")
            }
        }
    }

If I uncomment the testsources enabled:false line, it runs about as quickly as before. Normally our coverage run takes about 42 minutes. With the above configuration it's been running for about 3 hours and is about half finished. I attached YourKit to try and see where the time is going but something clover does seems to be confusing the profiler. All the method invocation times say 0ms which I've never seen before. I'll try profiling a different way to see if I can tell where the time is spent.

Hi Aaron, did you manage to install new SNAPSHOT version? Do you need any help? It would be great if you could give me a feedback till Friday the latest as I'm about to release the 3.1.11. Later feedback will be also very appreciated, but then fixes (if any) will go into the 3.1.12.

Created CLOV-1257.