C++ tree inlining bug with dynamic arrays

Jim Wilson wilson@cygnus.com
Mon Jan 24 22:06:00 GMT 2000

winsup (the cygwin library) fails to compile with the new C++ inlining on
trees code.  The problem is that winsup uses dynamic arrays.  Here is a
simple testcase

extern void __inline__
sub (int i)
  volatile char a[i];


sub2 ()
  sub (100);

In sub, DECL_SIZE (<VAR_DECL a>) is variable size, so it contains a SAVE_EXPR.
The SAVE_EXPR has a context that points at sub.  After inlining, the SAVE_EXPR
ends up in sub2, but the context still points at sub, triggering an abort
when we try to emit RTL for it.

cp/optimize.c has code to handle SAVE_EXPR contexts in remap_save_expr.
Unfortunately, this is only called or SAVE_EXPRs that exist in the function
body.  It does not get called for SAVE_EXPRs inside a VAR_DECL DECL_SIZE.

I believe that what needs to be done here is to add code to remap_decl
that will walk down the DECL_SIZE tree and copy it.  This patch does seem
to work.  However, I still get failures, and it turns out that at least one
more thing needs fixing:

Also, by looking at the var_decl, I see that 
is also wrong, but don't know if this will cause a crash.

Index: optimize.c
RCS file: /cvs/cvsfiles/devo/gcc/cp/optimize.c,v
retrieving revision 1.2
diff -p -r1.2 optimize.c
*** optimize.c	2000/01/03 19:18:42	1.2
--- optimize.c	2000/01/25 06:01:42
*************** remap_decl (decl, id)
*** 104,109 ****
--- 104,112 ----
        /* Make a copy of the variable or label.  */
        t = copy_decl_for_inlining (decl, fn, 
  				  VARRAY_TREE (id->fns, 0));
+       walk_tree (&DECL_SIZE (t), copy_body_r, id);
        /* Remember it, so that if we encounter this local entity
  	 again we can reuse this copy.  */
        n = splay_tree_insert (id->decl_map, 

More information about the Gcc-bugs mailing list