This bug has to do with the jdwp and hprof agents not parsing the
sun.boot.library.path (dll_dir) correctly since the fix for 6819213 went
in some years ago. This bug popped up while working on a particular
platform that does not have the ability to set LD_LIBRARY_PATH before
running the VM. As documented in the bug, on most platforms even if the
sun.boot.library.path consists of multiple paths and the jdwp or hprof
code fails to load a dependent lib, the system falls back to using
LD_LIBRARY_PATH so the failure is masked. On some other platforms, this
failover doesn't exist so we get an error when trying to load jdwp and
hprof dependent libs.
Some notes on a couple of files.
*
debugInit.c, hprof_init.c*:
Rearranged the init order so that the jvmti pointer gets initialized
before attempting to load the npt shared library.
*linker_md.c, hprof_md.c*
Much of the platform code in hprof and jdwp is duplicated and this
change is no different. Based on the code in hotspot
os_solaris/windows.cpp it parses the boot library path and attempts to
find the library.
*
error_messages.c*
Fixed a bug in the error message code that blindly dereferenced the npt
pointer even if it wasn't set because of an error in loading.
Webrev: http://cr.openjdk.java.net/~bpittore/7200297/webrev.00/
Ran the ute nsk/jdwp nsk/jvmti nsk/hprof tests, the JCK jdwp/jvmti tests
and some command line testing on windows and linux.
thanks,
bill
