[patch/rfc] kfail, not xfail descriptor parameter fails

[patch/rfc] kfail, not xfail descriptor parameter fails

[Andrew Cagney]

[-- Attachment #1: Type: text/plain, Size: 394 bytes --]

Hello,

GDB screws up the passing of function parameters when the ABI uses a 
descriptor (it passes the code instead of descriptor address). Anyway, 
it's a bug.  This changes the xfail's for that case into kfails 
(refining them a little - powerpc*-*-* is too broad, the tests work on 
some powerpc machines).

I think its almost but not quite obvious so I'll commit it in a day or so,
Andrew

[-- Attachment #2: diffs --]
[-- Type: text/plain, Size: 2750 bytes --]

2003-11-17  Andrew Cagney  <cagney@redhat.com>

	* gdb.base/callfuncs.exp: Change the XFAILed descriptor test to a
	KFAIL.

Index: gdb.base/callfuncs.exp
===================================================================
RCS file: /cvs/src/src/gdb/testsuite/gdb.base/callfuncs.exp,v
retrieving revision 1.11
diff -u -r1.11 callfuncs.exp
--- gdb.base/callfuncs.exp	3 Apr 2003 16:34:48 -0000	1.11
+++ gdb.base/callfuncs.exp	17 Nov 2003 22:53:27 -0000
@@ -182,30 +182,27 @@
     gdb_test "p t_func_values(func_val2,func_val1)" " = 0"
     gdb_test "p t_func_values(func_val1,func_val2)" " = 1"
 
-    # On the rs6000, we need to pass the address of the trampoline routine,
-    # not the address of add itself.  I don't know how to go from add to
-    # the address of the trampoline.  Similar problems exist on the HPPA,
-    # and in fact can present an unsolvable problem as the stubs may not
-    # even exist in the user's program.  We've slightly recoded t_func_values
-    # to avoid such problems in the common case.  This may or may not help
-    # the RS6000.
-    setup_xfail "rs6000*-*-*"
-    setup_xfail "powerpc*-*-*"
-    if {![istarget hppa*-*-hpux*]} then {
-	gdb_test "p t_func_values(add,func_val2)" " = 1"
-    }
-
-    setup_xfail "rs6000*-*-*"
-    setup_xfail "powerpc*-*-*"
-    if {![istarget hppa*-*-hpux*]} then {
-	gdb_test "p t_func_values(func_val1,doubleit)" " = 1"
-    }
-
-    setup_xfail "rs6000*-*-*"
-    setup_xfail "powerpc*-*-*"
-    if {![istarget hppa*-*-hpux*]} then {
-	gdb_test "p t_call_add(add,3,4)" " = 7"
-    }
+    # GDB currently screws up the passing of function parameters for
+    # ABIs that use function descriptors.  Instead of passing the
+    # address of te function descriptor, GDB passes the address of the
+    # function body.  This results in the called function treating the
+    # first few instructions of the function proper as a descriptor
+    # and attempting a jump through that (a totally random address).
+    setup_kfail "rs6000*-*-aix*" gdb/1457
+    setup_kfail "powerpc*-*-aix*" gdb/1457
+    setup_kfail "powerpc64*-*-*" gdb/1457
+    setup_kfail hppa*-*-hpux* gdb/1457
+    gdb_test "p t_func_values(add,func_val2)" " = 1"
+    setup_kfail "rs6000*-*-aix*" gdb/1457
+    setup_kfail "powerpc*-*-aix*" gdb/1457
+    setup_kfail "powerpc64*-*-*" gdb/1457
+    setup_kfail hppa*-*-hpux* gdb/1457
+    gdb_test "p t_func_values(func_val1,doubleit)" " = 1"
+    setup_kfail "rs6000*-*-aix*" gdb/1457
+    setup_kfail "powerpc*-*-aix*" gdb/1457
+    setup_kfail "powerpc64*-*-*" gdb/1457
+    setup_kfail hppa*-*-hpux* gdb/1457
+    gdb_test "p t_call_add(add,3,4)" " = 7"
     gdb_test "p t_call_add(func_val1,3,4)" " = 7"
 
     gdb_test "p t_enum_value1(enumval1)" " = 1"

Re: [patch/rfc] kfail, not xfail descriptor parameter fails

[Andrew Cagney]

> Hello,
> 
> GDB screws up the passing of function parameters when the ABI uses a descriptor (it passes the code instead of descriptor address). Anyway, it's a bug.  This changes the xfail's for that case into kfails (refining them a little - powerpc*-*-* is too broad, the tests work on some powerpc machines).
> 
> I think its almost but not quite obvious so I'll commit it in a day or so,

I've checked this in.

> 2003-11-17  Andrew Cagney  <cagney@redhat.com>
> 
> 	* gdb.base/callfuncs.exp: Change the XFAILed descriptor test to a
> 	KFAIL.

Andrew