[Bug fortran/34342] BOZ extensions not diagnosed as such with -std=f95

["burnus at gcc dot gnu dot org" <gcc-bugzilla at gcc dot gnu dot org>]

------- Comment #2 from burnus at gcc dot gnu dot org  2007-12-06 12:15 -------
Created an attachment (id=14702)
 --> (http://gcc.gnu.org/bugzilla/attachment.cgi?id=14702&action=view)
Draft patch for further BOZ diagnostic

The attached patch shows how one could implement:

- Error for initializing non-integers in DATA statements
  (invalid Fortran 2003/95)

- Print a warning for real(BOZ), dble(BOZ) etc., which are interpreted 
  different then required by Fortran 2003.

Missing:
- "REAL(BOZ)" does not print a warning, only "REAL(BOZ, kind)" works as
  for the former, gfc_simplify_real is not called. I need to find out what 
  route is taken for REAL() and INT().
- The "boz == true" needs to be "falsed" as some point, esp. after "int(BOZ)"
  it shall not be present.

(In principle, the warning could be replaced by a TRANSFER, which might be
enough to get Fortran 2003 BOZ working.)

Also missing:

a) As soon as there is Fortran 2003 BOZ support, one needs to modify
match_boz_constant. Currently, it gives with -std=f* an error if a BOZ is used
outside DATA; in Fortran 2003 this is allowed in REAL() etc., but not
elsewhere.

b) One may think of treating
  real_variable = BOZ
as
  real_variable = real(BOZ)
instead of as
  real_variable = int(BOZ)
But one has to think of how to handle
  real_variable = 0.0 + BOZ
  real_variable = 0 + BOZ
etc. in this case. I think other compilers tread only "lhs = BOZ" as real BOZ
and lhs = expression(BOZ) as integer BOZ, but one should do some tests to
confirm this and state clearly in
http://gcc.gnu.org/onlinedocs/gfortran/BOZ-literal-constants.html
how BOZ are treated.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=34342