[Bug fortran/41328] New: [4.4/4.5 regression] bad iostat when reading DOS file in a character array (non-advancing)

["Edouard dot Canot at irisa dot fr" <gcc-bugzilla at gcc dot gnu dot org>]

The following fortran code:

=== begin code ===
program main

   implicit none

   integer :: iostat, n_chars_read
   character(len=1) :: buffer(64) = ""

   open( unit=10, status="old", file="simple_dos.txt" )

   do

      read( unit=10, fmt='(64A)', advance='NO', iostat=iostat,          &
            size=n_chars_read ) buffer
      print *, ' n_chars_read = ', n_chars_read
      print *, ' buffer = "', buffer(1:n_chars_read), '"'

      if( is_iostat_eor(iostat) ) then
         print *, ' iostat = ', iostat, ' (END-OF-RECORD)'
      else if( is_iostat_end(iostat) ) then
         print *, ' iostat = ', iostat, ' (END-OF-FILE)'
         exit
      else
         print *, ' iostat = ', iostat
         stop 'ERROR'
      end if
      print *

   end do

end program main
=== end code ===

compiled in a standard way (just "-g" for debugging) cannot read more
than one line of a DOS file (which contain "\r\n" at the end of each
line).

For example, with the DOS file:

=== begin simple_dos.txt === (\r stands for the <CR> character)
a\r
b\r
c\r
=== end simple_dos.txt ===

it produces:

kaolin $ ./main
  n_chars_read =            1
  buffer = "a"
  iostat =           -2  (END-OF-RECORD)

  n_chars_read =            0
  buffer = ""
  iostat =         5008
STOP ERROR

When reading the UNIX version of the same file:

=== begin simple.txt ===
a
b
c
=== end simple.txt ===

it produces good results:

kaolin $ test_get_unit_2.8
  n_chars_read =            1
  buffer = "a"
  iostat =           -2  (END-OF-RECORD)

  n_chars_read =            1
  buffer = "b"
  iostat =           -2  (END-OF-RECORD)

  n_chars_read =            1
  buffer = "c"
  iostat =           -2  (END-OF-RECORD)

  n_chars_read =            0
  buffer = ""
  iostat =           -1  (END-OF-FILE)

______________________________________________________________

The bug doesn't arise when I replace the character array by an
ordinary character string, declared as:

   character(len=64) :: buffer = ""

This bug affects the 4.4.1 release and the 4.5 current beta, not
for 4.3.4, so we can considering a regression in 4.4/4.5

Moreover, valgind-3.5.0 detect some uninitialised value on line
245 of libgfortran/io/transfer.c, in the 'read_sf' routine of the
4.4/4.5 release.
Comparing this source file for the 4.3/4.4/4.5 versions, I saw that
the 'read_sf' has been completely redesigned after 4.3
When writing some information, I'm pretty sure that:
  *(p + 1) == '\n'
(inside the test on the line 245 of the same file) is not the valid
way to test the next character, because the pointer p points to only
one character, *(p + 1) always returns a NULL char.
Unfortunately, I cannot propose a patch.


-- 
           Summary: [4.4/4.5 regression] bad iostat when reading DOS file in
                    a character array (non-advancing)
           Product: gcc
           Version: 4.4.1
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: fortran
        AssignedTo: unassigned at gcc dot gnu dot org
        ReportedBy: Edouard dot Canot at irisa dot fr
 GCC build triplet: i686-linux-gnu
  GCC host triplet: i686-linux-gnu
GCC target triplet: i686-linux-gnu


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=41328