[Patch, Fortran] PR51578 Fix symbol import with renaming and only

[Tobias Burnus]

Dear all,

this email contains a updated patch, which replaces the FIXME by an 
implementation. See below.


Dear Dominique,

On 01/08/2012 11:13 PM, Dominique Dhumieres wrote:
> Your patch passed my tests and in top of that fixes the ICE for pr51522.f90.
:-)

> However I don't understand the errors:
>
>    integer(c_int) function vect_Utils_vuTriIneqHolds_c(u, v, tol, exception)&
>            1
> Error: Parameter 'c_int' at (1) has not been declared or is a variable, which does not reduce to a constant expression
>
>    real(c_double) function vect_Utils_vuNorm_c(u, tol, badLevel, exception) bind
>         1
> Error: Parameter 'c_double' at (1) has not been declared or is a variable, which does not reduce to a constant expression
>
> Is it related to the FIXME?

No, it's related to two bugs in the source code:
a) The USE statement is not a statement by its own (due to the tailing "&")
b) There is no USE statement for:  vect_Utils_vuNorm_c

I think the error message for (b) is good. I admit that for (a) a syntax 
error would have been better.

Excerpt from the test case (which has been created using "delta" which 
deletes lines, honouring certain constraints):

   integer(c_int) function vect_Utils_vuTriIneqHolds_c(u, v, tol, 
exception) &
     use, intrinsic :: iso_c_binding
   real(c_double) function vect_Utils_vuNorm_c(u, tol, badLevel, 
exception) bind(c)
   end function vect_Utils_vuNorm_c

* * *

Regarding the FIXME: No, it is completely unrelated and it is rather 
difficult to encounter.

Ingredients for reject-valid bugs related to the FIXME:
- One needs to USE a module name of an intrinsic module (iso_* or omp*)
- No non_intrinsic module with that name exists
- Multiple USE statements are used - some but not all with INTRINSIC 
attribute
- One renames symbols
- One uses the renamed-away symbols for other purposes

I believe that it is very unlikely that someone generates such a code.

Example:
   use iso_c_binding
   use, intrinsic :: iso_c_binding, c_double_orig => c_double
   integer :: c_double

That's valid and rejected. If a non_intrinsic module "iso_c_binding" 
exists, the code actually works correctly. It only wrongly 
use-associates "c_double" if no non_intrinsic module exists. Workaround: 
Use "intrinsic" in both USE statements - or in none.


However, I have now decided that it is better to actually implement the 
FIXME; the updated patch is attached. (It was simpler than expected.) 
Only gfc_use_modules has been changed and a new test case been added.

Build and regtested on x86-64-linux.
OK for the trunk?

Tobias
-------------- next part --------------
A non-text attachment was scrubbed...
Name: mod-use-v2.diff
Type: text/x-patch
Size: 20934 bytes
Desc: not available
URL: <http://gcc.gnu.org/pipermail/gcc-patches/attachments/20120109/73c6c463/attachment.bin>