Problem --with-gc=simple.

[Jeff Sturm]

On Wed, 25 Jul 2001, Antonio Ake wrote:
> java/lang/ConcreteProcess.java:0: Internal error: Segmentation fault

Thanks for the report.  It looks as though you discovered a latent bug in
the frontend.

The definition of BUILD_FILENAME_IDENTIFIER_NODE in java-tree.h cannot be
right, since ggc_mark_tree must only be called when a ggc_collect is in
progress.

This is what I think was intended:

Index: java-tree.h
===================================================================
RCS file: /cvs/gcc/gcc/gcc/java/java-tree.h,v
retrieving revision 1.97.2.14
diff -u -r1.97.2.14 java-tree.h
--- java-tree.h 2001/05/13 07:10:22     1.97.2.14
+++ java-tree.h 2001/07/25 22:51:46
@@ -1458,7 +1458,7 @@
   if (!((F) = maybe_get_identifier ((S))))     \
     {                                          \
       (F) = get_identifier ((S));              \
-      ggc_mark_tree ((F));                     \
+      ggc_add_tree_root ((F), 1);              \
     }
 
 /* Add a FIELD_DECL to RECORD_TYPE RTYPE.


> Is there any other way to disable to use mmap? Do you know who is using
> mmap, with gdb I got this:
> mmap<-__assert (?)
> mmap<-_OtsNotify_Ovf_I (?)

Why can't you use mmap?

Jeff