Community
Participate
Working Groups
Behavior is that running JUnit tests constrained by a subpackage (by either right clicking the package in the Package Explorer and running as JUnit or manually creating the Run profile) takes many minutes to run with the progress monitor displaying "Searching for errors in <project>" Further analysis during discussion of bug https://bugs.eclipse.org/bugs/show_bug.cgi?id=82852 [82852] reveals that excessive IO is occurring during creation of the JavaProject.newTypeHierarchy(..)
Created attachment 24447 [details] Stack trace of worker thread Trace shows the Type Hierarchy being created. Trace is over about 90 seconds of wall time.
Please read previous comments from bug https://bugs.eclipse.org/bugs/show_bug.cgi?id=82852 [82852] for further information. Project contains about 20k resources, with about 13k of them being .java files. Machine is WinXP, Dual Xeon 3.6ghz, 2gig. Eclipse is running on Sun Java 1.5_04 with mx heap of 768m.
Using region-based hierarchy to find all test cases just doesn't scale, as described in bug 82852. Why not instead using a standard search query with suitable search scope ?
Philippe, we are using a type hierarchy since we need to get all sub types of TestCase in a certain packages. I don't see how this can be done using the search engine. Can you please comment.
IType.newTypeHierarchy(...) should do the trick. You could constrain on a given project, and we could add an API using a searchscope to narrow the subtype search to a given set of elements (like a package here). The problem with a region is that it doesn't query indexes, and iterates all the types to connect them all. Then you likely only exploit a small subset of it. IRegion hierarchies got somewhat obsoleted once subtypes hierarchies got introduced.
Philippe, the call we are using is ITypeHierarchy newTypeHierarchy( IType type, IRegion region, IProgressMonitor monitor) throws JavaModelException; where the type is TestCase. So the type hierarchy is already constraint to subclasses of TestCase and a given region. So how different is this from creating a type hierarchy on TestCase can constrain it to a region. Instead of adding new API or use doing the filtering can't we change the method above to query/use the index since the "root" of the hierarchy is known here.
The method is defined on IJavaProject.
Created attachment 25418 [details] A patch that still relies on the region API but computes a more preceise region
Created attachment 25419 [details] Additional test cases
I have released the patches to HEAD. The final fix needs some further discussion with JDT/Core (the type hierarchy API on IType doesn't allow to constrain the hierarchy to a package).
Adam, can you please take the next integration build and see if the performance is better in your use case.
Dirk, I'm having a bit of trouble with the patch. I successfully applied the patch to my codebase and it indeed seems much much faster. However, it now appears to only 'shallowly' run tests in a package. It does not recursively go down package children looking for tests, it only executes ones on the absolute branch you select. So, for example, I have: test/com.tradecard.framework.EntityTest.java test/com.tradecard.framework.util.DBUnitPOCTest.java and I click the framework package, it will run EntityTest, but not the DBUnitPOCTest. I think perhaps you did too good of a job of constraining the package.
Oh, I should comment that I manually applied the patch supplied here, and did not take the integration build. So if this patch relies on API changes elsewhere in the codebase, perhaps that is why I am seeing this functionality as broken.
Adam, I tested it with 3.1 and there executing tests in a package didn't go down into "subpackages" as well (language wise something like a subpackage doesn't exist anyways). So in the example below: Test + p + Test1.java + p.q + Test2.java Executing JUnit tests on p only runs Test1 (I tested it with 3.1). Can you provide a test case (or a workspace) where subpackages are considered ?
You know what, you are correct. I have not been running JUnit tests through Eclipse for so long because of this bug I didn't realize that was the old functionality. However, I would make an argument that the 'subpackage' recursive functionality is correct and what people would expect. I know that if I use an ant task to run JUnit tests and specify a package, I will get the recursive functionality. But, perhaps this is a new bug.
The sub package bug is https://bugs.eclipse.org/bugs/show_bug.cgi?id=46433. Adam, the patch went into yesterdays intergration build and if we are happy with it I will backport it to 3.1.1.
Markus, Tobias can you please have a look at the patch and review it. Since the preformance improvements are significant I would like to back port it to 3.1.1. The code is basically borrowed from TypeHierarchyLifeCycle
Created attachment 26525 [details] An improved patch
Patch looks good to me, +1 for 3.1.1
I will mark this bug as fixed for 3.1.1 Opened bug 108254 and 108255 to track this issue in the 3.2 stream.
Start verifying...
Verified in M20050831-1200 + ZRH plugins from R3_1_maintenance.
