970125-0.f

[Jeffrey A Law <law at cygnus dot com>]

I think the problem with this testcase is actually in the Fortran
front-end, but not being an expert on trees, I'm not really sure.

Basically it looks like when building an argument list for a
function call we end up creating this kind of tree for one
of the arguments:

(gdb) p debug_tree(item)
 <addr_expr 0x82b8c70
    type <pointer_type 0x82cbe5c
        type <pointer_type 0x82cbd94 type <integer_type 0x82c1dbc integer>
            allocated from permanent_obstack
            unsigned permanent SI
            size <integer_cst 0x82b6fcc constant permanent 32>
            align 32 symtab 0
            pointer_to_this <pointer_type 0x82cbe5c>>
        allocated from permanent_obstack
        unsigned permanent SI size <integer_cst 0x82b6fcc 32>
        align 32 symtab 0>
    allocated from momentary_obstack
   
    arg 0 <addr_expr 0x82b8c34 type <pointer_type 0x82cbd94>
        allocated from momentary_obstack
       
        arg 0 <var_decl 0x82b9fc0 i4 type <integer_type 0x82c1dbc integer>
            allocated from function maybepermanent obstack
            addressable used common SI file k.f line 1 size <integer_cst 0x82b6fcc 32>
            align 32 context <function_decl 0x82cbbb8 MAIN__>
            (mem:SI (addressof:SI (reg:SI 23) 21))>>>
 


Note that we have (ADDR_EXPR (ADDR_EXPR ...))  which seems more than
a little strange to me.


This happens in ffecom_arg_ptr_to_expr at this line:

13222
13223         item = ffecom_1 (ADDR_EXPR, build_pointer_type (TREE_TYPE (item)),
13224                          item);
13225         return item;



When we get to this point item is:

 <addr_expr 0x82b8c34
    type <pointer_type 0x82cbd94
        type <integer_type 0x82c1dbc integer allocated from permanent_obstack
            permanent SI
            size <integer_cst 0x82b6fcc constant permanent 32>
            align 32 symtab 0 precision 32
            min <integer_cst 0x82c1e18 constant permanent -2147483648>
            max <integer_cst 0x82c1e30 constant permanent 2147483647>
            pointer_to_this <pointer_type 0x82cbd94>>
        allocated from permanent_obstack
        unsigned permanent SI size <integer_cst 0x82b6fcc 32>
        align 32 symtab 0>
    allocated from momentary_obstack
   
    arg 0 <var_decl 0x82b9fc0 i4 type <integer_type 0x82c1dbc integer>
        allocated from function maybepermanent obstack
        addressable used common SI file k.f line 1 size <integer_cst 0x82b6fcc 32>
        align 32 context <function_decl 0x82cbbb8 MAIN__>
        (mem:SI (addressof:SI (reg:SI 23) 21))>>


Does it make sense to make (ADDR_EXPR (ADDR_EXPR ...)??


If it doesn't, then the bug is the Fortran front end.


jeff