Community
Participate
Working Groups
Test failures JDT UI in build N20101105-2000 http://download.eclipse.org/eclipse/downloads/drops/N20101105-2000/testresults/xml/org.eclipse.jdt.ui.tests_win32.win32.x86_6.0.xml In the following snippet value of 'a' is not used, but the compiler does not report any warning. package test1; public class E { private void foo() { int a= 10; a++; a--; --a; ++a; for ( ; ; a++) { } } }
Stephan, I am assigning it to you. Let me know if you are willing to work on it. Thanks.
I'd be happy to fix this, but on my machine it ain't broken. More specifically: my base is a fresh SDK 3.7M3, workspace contains - jdt.core v_B22 (which was also used in N20101105-2000) - jdt.ui.tests HEAD - and a few more Running LocalCorrectionsQuickFixTest yields 100% pass (161/161). Moreover, according to the map files the following builds have not differences in jdt land: I20101102-0800 (no test failures) N20101105-2000 (failures in LocalCorrectionsQuickFixTest) Simplest way to create the bug is by deleting these lines in tests: JavaCore.setOptions(hashtable); Thus, it *might* be another hint that setting JavaCore options has a rarely triggered racing condition (cf. bug 302850).
(In reply to comment #2) > I'd be happy to fix this, but on my machine it ain't broken. > > More specifically: my base is a fresh SDK 3.7M3, workspace contains > - jdt.core v_B22 (which was also used in N20101105-2000) > - jdt.ui.tests HEAD > - and a few more > > Running LocalCorrectionsQuickFixTest yields 100% pass (161/161). > > Moreover, according to the map files the following builds have not differences > in jdt land: > I20101102-0800 (no test failures) > N20101105-2000 (failures in LocalCorrectionsQuickFixTest) Nightly builds are run against the latest code in HEAD. Only Integration builds depend on map files. (http://download.eclipse.org/eclipse/downloads/build_types.html)
(In reply to comment #3) > Nightly builds are run against the latest code in HEAD. Only Integration builds > depend on map files. > (http://download.eclipse.org/eclipse/downloads/build_types.html) Sorry, yes (I was just misled by the mapfile link on the download page). Indeed the failures can only be reproduced with the patch from bug 328519 comment 11. HEAD is not affected currently, because Olivier reverted that patch.
I think we can close this one as being a side-effect of a wrong fix for bug 328519. I need to work more on bug 328519 to support the case where we tag a variable as USED only because we have preserveAllLocals option enabled. Even if this option is enabled, we should still generate the warning. Closing as INVALID
Verified for 3.7M4.