I diffed woody's gcc-<version>/gcc-<version>.tar.bz2/libjava/java/lang/natString.cc with sarge's, and noticed that in java::lang::String::rehash(), two calls to _Jv_AllocBytesChecked were replaced by calls to _Jv_AllocBytes (this was the only change in that function).
Because the deadlock/livelock happend at a power of 2 (a roommate of mine could reproduce it at 2^14 with gcj 3.4.1 on Mandrake on powerpc (gcc-java-3.4.1-4mdk)), this is a very suspecious change, since rehashing is generally done at powers-of-2. --Jeroen -- Jeroen van Wolffelaar [EMAIL PROTECTED] http://jeroen.A-Eskwadraat.nl