[Bug fortran/69055] Internal compiler error -fsanitize=float-cast-overflow

["sgk at troutmask dot apl.washington.edu" <gcc-bugzilla at gcc dot gnu dot org>]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=69055

--- Comment #3 from Steve Kargl <sgk at troutmask dot apl.washington.edu> ---
On Sat, Dec 26, 2015 at 05:43:56PM +0000, sgk at troutmask dot
apl.washington.edu wrote:
> https://gcc.gnu.org/bugzilla/show_bug.cgi?id=69055
> 
> --- Comment #2 from Steve Kargl <sgk at troutmask dot apl.washington.edu> ---
> On Sat, Dec 26, 2015 at 05:15:27PM +0000, kargl at gcc dot gnu.org wrote:
> > > 
> > >        N=daten(j)%daten
> > >
> > 
> > The code is non-conforming Fortran.  daten(j)%daten is used uninitialized.
> > gfortan can do anything it wants with the code, including throwing an 
> > error.
> > 
> 
> After making the code at least conform to the Fortran 
> standard and removing the overuse of "daten", the error
> still occurs.
> 
> subroutine tstd
>    implicit none
>    integer n
>    type a
>       real(8) b
>    end type a
>    type(a) :: daten(10)
>    integer ::  j = 1
>    daten(:)%b = 42
>    n = daten(j)%b
> end subroutine tstd
> 
> The -fdump-tree-original output is 
> 
> tstd ()
> {
>   struct a daten[10];
>   static integer(kind=4) j = 1;
>   integer(kind=4) n;
> 
>   {
>     integer(kind=4) S.0;
> 
>     S.0 = 1;
>     while (1)
>       {
>         if (S.0 > 10) goto L.1;
>         daten[S.0 + -1].b = 4.2e+1;
>         S.0 = S.0 + 1;
>       }
>     L.1:;
>   }
>   n = (integer(kind=4)) daten[NON_LVALUE_EXPR <j> + -1].b;
> }
> 
> which looks like what I expect.  It seems that -fsanitize=float-cast-overflow
> option has a false positive.
> 

Here's the backtrace.

Program received signal SIGSEGV, Segmentation fault.
0x08a9852b in build_call_expr_loc_array(unsigned int, tree_node*, int,
tree_nod**) ()
(gdb) bt
#0  0x08a9852b in build_call_expr_loc_array(unsigned int, tree_node*, int,
treenode**) ()
#1  0x08a986d0 in build_call_expr_loc(unsigned int, tree_node*, int, ...) ()
#2  0x08808f7a in ubsan_instrument_float_cast (loc=<optimized out>, 
    type=<optimized out>, expr=<optimized out>, arg=<optimized out>)
    at /mnt/kargl/gcc/gcc/ubsan.c:1609
#3  0x083df160 in convert_to_integer_1 (type=0x296533c0, expr=0x2a10d654, 
    dofold=<optimized out>) at /mnt/kargl/gcc/gcc/convert.c:923
#4  0x083de656 in convert_to_integer (type=0x296533c0, expr=0x2a114214)
    at /mnt/kargl/gcc/gcc/convert.c:968
#5  0x08293b1f in convert (type=<optimized out>, expr=0x2a114214)
    at /mnt/kargl/gcc/gcc/fortran/convert.c:101
#6  0x08309ce8 in gfc_conv_intrinsic_conversion (se=0xbfbfe318, 
    expr=<optimized out>) at /mnt/kargl/gcc/gcc/fortran/trans-intrinsic.c:336
#7  0x082ff1cb in gfc_conv_intrinsic_function (se=0xbfbfe318, expr=0x29e13690)
    at /mnt/kargl/gcc/gcc/fortran/trans-intrinsic.c:7877
#8  0x082fad05 in gfc_conv_function_expr (se=0xbfbfe318, expr=0x29e13690)
    at /mnt/kargl/gcc/gcc/fortran/trans-expr.c:6527
#9  0x082e2a4f in gfc_conv_expr (se=0xbfbfe318, expr=0x29e13690)
    at /mnt/kargl/gcc/gcc/fortran/trans-expr.c:7522
#10 0x082eac96 in gfc_trans_assignment_1 (expr1=0x29e13620, 
    expr2=<optimized out>, init_flag=<optimized out>, dealloc=<optimized out>)
    at /mnt/kargl/gcc/gcc/fortran/trans-expr.c:9249
#11 gfc_trans_assignment (expr1=0x29e13620, expr2=0x29e13690, init_flag=false, 
    dealloc=<optimized out>) at /mnt/kargl/gcc/gcc/fortran/trans-expr.c:9485
#12 0x082fb64b in gfc_trans_assign (code=0x29e13770)
    at /mnt/kargl/gcc/gcc/fortran/trans-expr.c:9497
#13 0x082af9b2 in trans_code (code=<optimized out>, cond=0x0)
    at /mnt/kargl/gcc/gcc/fortran/trans.c:1660
#14 0x082dd58c in gfc_generate_function_code (ns=<optimized out>)
    at /mnt/kargl/gcc/gcc/fortran/trans-decl.c:6083
#15 0x0824737f in translate_all_program_units (
    gfc_global_ns_list=<optimized out>)
    at /mnt/kargl/gcc/gcc/fortran/parse.c:5612
#16 gfc_parse_file () at /mnt/kargl/gcc/gcc/fortran/parse.c:5818
#17 0x082974b8 in gfc_be_parse_file ()
    at /mnt/kargl/gcc/gcc/fortran/f95-lang.c:201
#18 0x087e4e38 in compile_file () at /mnt/kargl/gcc/gcc/toplev.c:464
#19 0x087e4a9d in do_compile () at /mnt/kargl/gcc/gcc/toplev.c:1977
#20 toplev::main (this=<optimized out>, argc=<optimized out>, 
    argv=<optimized out>) at /mnt/kargl/gcc/gcc/toplev.c:2084
#21 0x08fcd3ee in main (argc=<optimized out>, argv=<optimized out>)
    at /mnt/kargl/gcc/gcc/main.c:39

Note, using an explicit Fortran conversion function, avoids the ICE.

subroutine tstd
   implicit none
   integer n
   type a
      real(8) b
   end type a
   type(a) :: daten(10)
   integer ::  j = 1
   daten(:)%b = 42
   n = int(daten(j)%b)
end subroutine tstd

Note, the -ftree-dump-original is identical to the above dump!
Looks like -fsanitize=float-cast-overflow does not understand
Fortran mix-mode math semantics.