Internal compiler bug using an array of integer(kind=1)

[Steve Kargl]

On Fri, Oct 10, 2008 at 09:28:06AM +0200, Tobias Burnus wrote:
> Hi Sebastian,
> 
> Steve Kargl wrote:
> > On Fri, Oct 10, 2008 at 12:09:37AM +0200, Sebastian Gallinat wrote:
> >   
> >> c:\silverbox\archiv\lib\libharu\gfortran\demo\mixed>gfortran bug.f95 -o 
> >> bug.f95:1: internal compiler error: in gfc_conv_array_initializer, at
> >>     
> Thanks for the bug report, I opened
> http://gcc.gnu.org/bugzilla/show_bug.cgi?id=37792.
> 

Given the reduced testcase

  program bug
  implicit none
  integer(1) :: a(2) = (/ Z'FF', Z'FF' /)
  print*, a
  end program bug

The attached patch gives

troutmask:sgk[216] gfc4x -o z a.f90
a.f90:3.23:

integer(1) :: a(2) = (/ Z'FF', Z'FF' /)
                      1
Error: Arithmetic overflow converting INTEGER(16) to INTEGER(1) at (1).
This check can be disabled with the option -fno-range-check

troutmask:sgk[217] gfc4x -o z -fno-range-check a.f90
troutmask:sgk[218] ./z
   -1   -1

The regession test reports

                === gfortran Summary ===

# of expected passes            28266
# of unexpected successes       23
# of expected failures          27
# of unsupported tests          42


2008-10-10  Steven G. Kargl  <kargls@comcast.net>

	* trans-array.c (gfc_conv_array_initializer):  Overflowing BOZ conversion
	can interfere with array initialization expressions.

PS: I don't have commit access.
-- 
Steve
-------------- next part --------------
A non-text attachment was scrubbed...
Name: abc.diff
Type: text/x-diff
Size: 733 bytes
Desc: not available
URL: <http://gcc.gnu.org/pipermail/fortran/attachments/20081011/ffbfe89a/attachment.bin>