GIJ segfaults on java-puzzlers/5-exceptional-puzzlers/puzzle-45/Workout.java from Bloch and Gafter's latest book. (This "puzzle" is due to Jim Hugunin of Jython and IronPython fame.)
| $ curl -O http://www.javapuzzlers.com/java-puzzlers.zip | % Total % Received % Xferd Average Speed Time Time Time Current | Dload Upload Total Spent Left Speed | 100 61564 100 61564 0 0 37113 0 0:00:01 0:00:01 --:--:-- 43957 | $ unzip -q java-puzzlers.zip | $ gcj -C java-puzzlers/5-exceptional-puzzlers/puzzle-45/Workout.java | $ gij -classpath java-puzzlers/5-exceptional-puzzlers/puzzle-45 Workout | Segmentation fault | $ gcj --main=Workout java-puzzlers/5-exceptional-puzzlers/puzzle-45/Workout.java | $ ./a.out | Segmentation fault The above is under the following version of GCJ: | $ gcj --version | gcj (GCC) 4.0.1 20050727 (Red Hat 4.0.1-5) | Copyright (C) 2005 Free Software Foundation, Inc. | This is free software; see the source for copying conditions. There is NO | warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. I also tried running this under three other 32-bit VMs: Sun 1.4.2_08, BEA (build 1.4.2_05-b04), and IBM 1.4.2 (build cxia32142-20050609). The only one that seems to run this program correctly is Sun's JVM. (Given a limited maximum stack depth, the correct behavior for this program is to run a few gazillion years and terminate successfully. In Sun's case, the maximum stack depth is 1024 and the program can be expected to run for about 10^300 years, give or take a few orders of magnitude.) IBM's 32-bit Linux JDK starts spewing out the following diagnostic: | Stack overflow occurred outside JITted code or Interpreter. | Searching for valid stack frames. | Stack trace information below may not be accurate or complete. | It is provided for diagnostic purposes. | Stack overflow occurred outside JITted code or Interpreter. | Searching for valid stack frames. | Stack trace information below may not be accurate or complete. | It is provided for diagnostic purposes. | at Workout.workHard | Stack overflow occurred outside JITted code or Interpreter. | Searching for valid stack frames. | Stack trace information below may not be accurate or complete. | It is provided for diagnostic purposes. | at Workout.workHard I had to Ctrl-C out of it after it generated 100,000+ lines of the above. BEA's JDK seems to do the right thing: run till the end of times. On closer inspection, it turns out to get a little hosed: it stops responding to SIGTERM and SIGQUIT. Bryce pointed out to me that Sun's behavior is not the only correct interpretation of the above program's semantics. If the runtime (or the native ahead-of-time compiler) performs tail-call optimization, then the program may loop forever without ever terminating. After thinking about this over the weekend, I came to realize that this is wrong. There seems to be no room for TCO in this particular case: private static void workHard() { try { workHard(); } finally { workHard(); } } Note that while the recursive call in the "finally" block *is* in the tail position, the call in the "try" block is not. Therefore, the recursion cannot be transformed into a loop. -- Summary: GCJ/GIJ segfaults on Workout.java from Click and Hack's "Java Puzzlers" Product: gcc Version: 4.0.1 Status: UNCONFIRMED Severity: normal Priority: P3 Component: java AssignedTo: unassigned at gcc dot gnu dot org ReportedBy: vadimn at redhat dot com http://gcc.gnu.org/bugzilla/show_bug.cgi?id=24980
