The VM calls IdealGraphPrinter::cleanup() at the final safepoint as the VM terminates, to end the XML stream. This seems to happen with no problem when termination is due to the last non-daemon thread exiting.

The cleanup() call is missing when the VM terminates via System.exit. When we tried to add that as part of JDK-8222534:

@@ -467,10 +482,18 @@
 
   set_vm_exited();
 
+ // Clean up ideal graph printers - Compiler threads will be
+ // blocked by the current safepoint.
+#if defined(COMPILER2) && !defined(PRODUCT)
+ IdealGraphPrinter::clean_up();
+#endif
+
   // cleanup globals resources before exiting. exit_globals() currently
   // cleans up outputStream resources and PerfMemory resources.
   exit_globals();

we encountered a crash when running with:

-XX:PrintIdealGraphFile=tmp.xml -XX:+PrintIdealGraph -XX:PrintIdealGraphLevel=4

# Internal Error (open/src/hotspot/share/utilities/xmlstream.cpp:184), pid=32637, tid=32644
# assert((!inside_attrs()) || VMError::is_error_reported()) failed: cannot close element inside attrs
#

It seems a compiler thread was already in the middle of writing a chunk of XML when it was stopped by the terminal safepoint. Examining tmp.xml we see the final entry is:

<method

which traces back to

void IdealGraphPrinter::print_method(ciMethod *method, int bci, InlineTree *tree) {
  begin_head(METHOD_ELEMENT);

  stringStream str;
  method->print_name(&str);

and we then find the problem in:

void ciMethod::print_name(outputStream* st) {
  check_is_loaded();
  GUARDED_VM_ENTRY(get_Method()->print_name(st);)
}

as the guarded entry can lead to safepoint blocking.

The problem should be reproducible in the last non-daemon thread termination case if we have a daemon thread alive that is still triggering compilation.