[PATCH] Fix for PR java/9577: Name Mangling Error

[Ranjit Mathew <rmathew at hotmail dot com>]

Hi,

    Here is a proposed patch for PR java/9577 (reported by 
yours truly). To summarise, we were not compressing the
"java" in a "java.lang.String" in the mangled method
name, that appears as a method parameter after a
"java.io.File[]", while the reverse worked. The C++
mangler however, correctly mangled the method name
in both the cases leading to a linker error with
CNI code in the former case.

The problem was that find_compression_record_match( )
incorrectly assumes that all IDENTIFIER_NODEs are
partial package names in the compression table - a
"6JArray" that is emitted for an array parameter
breaks this assumption and causes the reported bug.

Tested with a clean libgcj build on i686-pc-linux-gnu
with a GCC 3.4 snapshot (20030806) and this patch
did not cause any unexpected failures in the libgcj
testsuite (without Mauve and Jacks).

The patch also fixes a couple of typos in comments
elsewhere in this file (mangle.c).

Ranjit.


Index: ChangeLog
from  Ranjit Mathew  <rmathew@hotmail.com>

	Fixes PR java/9577
	* mangle.c (find_compression_record_match): Skip
	over a "6JArray" (the array template mangled string)
	IDENTIFIER_NODE.
	(mangle_array_type): Correct minor typo.
	(atms): Move definition to the beginning.

Index: mangle.c
===================================================================
--- mangle.c	2003-08-09 13:44:52.000000000 +0530
+++ mangle.c	2003-08-10 15:28:17.000000000 +0530
@@ -68,4 +68,7 @@ struct obstack *mangle_obstack;
   obstack_grow (mangle_obstack, (S), sizeof (S)-1)
 
+/* atms: array template mangled string. */
+static GTY(()) tree atms;
+
 /* This is the mangling interface: a decl, a class field (.class) and
    the vtable. */
@@ -234,5 +237,5 @@ mangle_type (tree type)
    would generate three entries: two package names and a type. If
    java.lang.String is presented next, the java.lang will be matched
-   against the first two entries (and kept for compression as S_0), and
+   against the first two entries (and kept for compression as S0_), and
    type String would be added to the table. See mangle_record_type.
    COMPRESSION_NEXT is the index to the location of the next insertion
@@ -302,10 +305,14 @@ find_compression_record_match (tree type
 	    break;
 	  }
+        else if (atms && TREE_VEC_ELT (compression_table, j) == atms)
+          {
+            /* Skip over a "6JArray". */
+          }
 	else
 	  {
 	    /* We don't want to match an element that appears in the middle
 	    of a package name, so skip forward to the next complete type name.
-	    IDENTIFIER_NODEs are partial package names while RECORD_TYPEs
-	    represent complete type names. */
+	    IDENTIFIER_NODEs (except for a "6JArray") are partial package
+            names while RECORD_TYPEs represent complete type names. */
 	    while (j < compression_next 
 		   && TREE_CODE (TREE_VEC_ELT (compression_table, j)) == 
@@ -414,9 +421,7 @@ mangle_pointer_type (tree type)
 /* Mangle an array type. Search for an easy solution first, then go
    through the process of finding out whether the bare array type or even
-   the template indicator where already used an compress appropriately.
+   the template indicator were already used and compressed appropriately.
    It handles pointers. */
 
-/* atms: array template mangled string. */
-static GTY(()) tree atms;
 static void
 mangle_array_type (tree p_type)
@@ -437,5 +442,5 @@ mangle_array_type (tree p_type)
     }
 
-  /* Maybe we have what we're looking in the compression table. */
+  /* Maybe we have what we're looking for in the compression table. */
   if ((match = find_compression_array_match (p_type)) >= 0)
     {