Re: [tree-ssa] Lowering of VLA's, part 2

[law at redhat dot com]

In message <20030827211526.GA18605@atrey.karlin.mff.cuni.cz>, Zdenek Dvorak wri
tes:
 >Hello,
 >
 >> >> >> > 	* gimplify.c (struct tree_p_list): New type.
 >> >> >> > 	(join_tree_p_list, add_tree_p_list, free_tree_p_list,
 >> >> >> > 	build_stack_save_restore, prepend_to_exit_gotos): New functions
 >.
 >> >> >> 
 >> >> >> Use a TRY_FINALLY_EXPR instead.
 >> >> >
 >> >> > unnecesarily creating new eh handling constructs (that we really do no
 >t
 >> >> > manage somehow greatly) seems a bit like overkill to me.
 >> >
 >> >> If we want the stack to be restored properly when we
 >> >> exit a block with a throw, we need to use an EH construct.
 >> >
 >> > do we indeed want this? Of course it won't spoil anything, but
 >> > we can live without it.
 >> 
 >> We absolutely want this.  That we don't do it currently is a bug.
 >
 >not really.  It does not change correctness, and since exceptions
 >should not occur frequently, the performance/temporary space leak tradeoff
 >might be justified.
Err, I thought deallocating as we left the scope was mandated by the
standard.  At least I thought someone sent me verbage from the C99
standard to that effect when I suggested not deallocating until the
function exited....

Jeff