Community
Participate
Working Groups
When I build Eclipse JDT using OpenJDK, I get a number of bugs [javac] /org.eclipse.jdt.debug/jdi/org/eclipse/jdi/internal/request/EventRequestManagerImpl.java:432: error: inconvertible types [javac] return new ArrayList<AccessWatchpointRequest>((Collection<? extends AccessWatchpointRequest>) fRequests.get(ACCESS_WATCHPOINT_INDEX)); [javac] ^ [javac] required: Collection<? extends AccessWatchpointRequest> [javac] found: HashSet<EventRequest> Initially I thought it is a bug in OpenJDK compiler, but after closer investigation it looks like JDT allows for potentially unsafe code. To understand the problem, lets look at the offending line: return new ArrayList<AccessWatchpointRequest>((Collection<? extends AccessWatchpointRequest>) fRequests.get(ACCESS_WATCHPOINT_INDEX)); and field declaration private ArrayList<HashSet<EventRequest>> fRequests; fRequests.get(ACCESS_WATCHPOINT_INDEX) returns a HashSet<EventRequest>, which may or may not be casted to (Collection<? extends AccessWatchpointRequest>), because EventRequest is a parent of AccessWatchpointRequest. Compiler cannot know that this code will work.
See https://bugs.eclipse.org/bugs/show_bug.cgi?id=205195#c7 There are many instances where javac implementation is tighter than the underspecified rules in this area. We don't have plans to match (the undocumented) behavior here.
@Daniel: From your blog entry I understood that you are patching this. Could you provide a patch for this specific situation so that Sankaran could check if it would be save to apply to JDT. I think it would be nice to be able to compile Eclipse with OpenJDK, that is the de-facto standard in the Javac world.
(In reply to comment #2) > @Daniel: From your blog entry I understood that you are patching this. Could > you provide a patch for this specific situation so that Sankaran could check if > it would be save to apply to JDT. What is being patched ? Debug or the compiler ? The latter approach could be problematic as some times when we attempted to match javac on specific cases in the past, it backfired in some other cases and we ended up in a worse situation (i.e we could not compile something javac would which is much worse than our accepting something javac would not)
@Sankaran I believe Daniel has a patch for the source code, e.g. EventRequestManagerImpl.java. Maybe I'm wrong, lets see what Daniel says.
I have seen some inconsistency between ecj and javac earlier, usually ecj was correct. In Eclipse Fedora Build, I patch the failing source. The patch is available here: http://git.eclipse.org/c/linuxtools/org.eclipse.linuxtools.eclipse-build.git/tree/eclipse-build/patches/jdt-debug-compilation.patch. Unfortunately the differences between javac and ecj are big enough to break ecj compiler when the patch is applied (unnecessary casts, unchecked exceptions etc). Frankly speaking, I do not have an idea how to adjust source code to make it working for both ecj and javac and do not introduce too many changes... -- Chris
Why don't give JDT/Debug team a chance to look at this first ? Recently things got generified there.
@Chris: Thanks. I'm surprised how small the problem is. Thats all what is preventing Eclipse from getting compiled by OpenJDK?
Yes, that's the whole patch - and I was able to build Eclipse with it : http://koji.fedoraproject.org/koji/taskinfo?taskID=3739589
(In reply to comment #8) > Yes, that's the whole patch - and I was able to build Eclipse with it : > http://koji.fedoraproject.org/koji/taskinfo?taskID=3739589 The patch introduces 32 compile errors in my workspace - 16 unnecessary cast errors and 16 type safety errors. Description Resource Path Location Type Type safety: Unchecked cast from Object to Collection<? extends AccessWatchpointRequest> EventRequestManagerImpl.java /org.eclipse.jdt.debug/jdi/org/eclipse/jdi/internal/request line 432 Java Problem
(In reply to comment #9) > The patch introduces 32 compile errors in my workspace - 16 unnecessary cast > errors and 16 type safety errors. Chatting a bit with Markus there does not appear to be a good type-safe way to fix this, other than to separate out the lists of EventRequests. I'll code something up today.
Markus had a great idea of using a generic type to separate out the collections and avoid casting / type safety problems. So building on his example, I pushed the changes to http://git.eclipse.org/c/jdt/eclipse.jdt.debug.git/commit/?id=36ba3d537eb874455538949bb1fb509d4544188f Krzysztof, please verify.
I was able to built org.eclipse.jdt.debug (3.7.100.v20120201-1716) with javac. It looks OK.
Thanks Chris, Srikanth and Michael. Great to hear! Thanks for your efforts.
