This is the mail archive of the gcc@gcc.gnu.org mailing list for the GCC project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]

[PATCH] __builtin_apply / -fdefer-pop bug (ObjC fix!)


(please CC: to me, as I do not follow the *@gcc.gnu.org lists)

	As many well know, GCC since 2.8.x has had a lot
of bugs with the __builtin_apply() function, especially
with the ObjC front-end. Come to find out, it doesn't
appear to be a problem with __builtin_apply() at all,
but with an over-optimization of -fdefer-pop.

	When compiled with -fdefer-pop (without the following
patch) GCC neglects to set the stack pointer appropriately
after the memcpy() call (to move the stack arguments), and
the called function (that __builtin_apply() was trying to call)
is sent a bogus stack pointer.

	With the following patch, a NO_DEFER_POP/OK_DEFER_POP
pair - which temporarily inhibits -fdefer-pop - is placed 
around the critical section in:

	gcc/expr.c:expand_builtin_apply()

	Upon recompile of libobjc (or any other source that uses
__builtin_apply()) correct functionality of __builtin_apply
(and henve forward::/performv:: ) is restored to the ObjectiveC
language runtime.

	A closer inspection as to why the -fdefer-pop optimization
fails in this instance should be looked at, but this workaround
appears to mirror similar uses of NO_DEFER_POP/OK_DEFER_POP in
expr.c.

	* gcc/expr.c - Fix expand_builin_apply() due to -fdefer-pop breakage

------------------- Cut here ----------------
--- expr.c.old	Wed Jun 30 18:59:55 1999
+++ expr.c	Wed Jul 26 16:21:37 2000
@@ -9987,6 +9987,8 @@
   rtx old_stack_level = 0;
   rtx call_fusage = 0;
 
+  NO_DEFER_POP;
+
   /* Create a block where the return registers can be saved.  */
   result = assign_stack_local (BLKmode, apply_result_size (), -1);
 
@@ -10145,6 +10147,8 @@
   else
 #endif
     emit_stack_restore (SAVE_BLOCK, old_stack_level, NULL_RTX);
+
+  OK_DEFER_POP;
 
   /* Return the address of the result block.  */
   return copy_addr_to_reg (XEXP (result, 0));
------------------- Cut here ----------------
-- 
Jason McMullan, Senior Linux Consultant, Linuxcare, Inc.
412.422.8077 tel, 412.656.3519 cell, 415.701.0792 fax
jmcmullan@linuxcare.com, http://www.linuxcare.com/
Linuxcare. Support for the revolution.

Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]