[RFC, Patch, Fortran] F2003's OOP constructor support

[Tobias Burnus]

Dear all,

attached is a draft patch for Fortran 2003's constructors. By 
constructors, I mean a generic function which has the same name as a 
derived type and which can be used in place of a structure constructor.

If the generic function returns the derived type, it resembles the 
constructor functions of C++; however, the generic function might also 
just return an integer and have no relation/knowledge of the derived 
type. (Only) if no suitable generic function is available, the structure 
constructor is used.


The attached patch is nearly ready. Thus, I would like to have feedback 
on the implementation. Additionally, I would be happy if someone could 
test it as the patch could potentially break any Fortran code which uses 
derived types - and it might not handle all valid constructor code 
either. Especially with invalid code there can be issues.


The patch is not yet cleaned up; I also need to write a changelog. For 
the status and possibly updated versions of the patch, see also 
https://userpage.physik.fu-berlin.de/~tburnus/tmp/constructor.diff

While the patch should have no major regressions nor missing features, 
it is not yet regression free. Currently known issues (see link above 
for details):
- gfortran.dg/class_21.f03: ICE on valid code.
- gfortran.dg/coarray_lock_3.f90 and coarray_lock_4.f90: The dg-error 
message changed.
- fortran.dg/result_1.f90: Ditto.
- constructor_4.f90: (new test) Missing -std=f95 diagnostic
- [no test] Missing ambiguous check
- (Known, unrelated failures: realloc_on_assign_5.f03, 
select_type_12.f03, entry_4.f90)


Internal implementation:
* For derived types, a generic-function symbol is generated
* Additionally, a symbol is generated for FL_DERIVED with the same 
sym->name but where the symtree->name has the first letter of the name 
uppercased. This is included in the sym->generic list of previous symbol.
* At resolve time, if not suitable generic function could be found, a 
structure constructor is assumed.
* The main issue is to have FL_DERIVED available where needed and to 
ensure that with module loading always both symbols are loaded.
* I had to modify iso_c_binding a bit, which resulted in fixing PR 37829 
and 45190.

Tobias

PS: I have tested the patch with some real-word codes, though not all 
with the latest patch version. I do not expect major issues, but I do 
expect that some programs will fail. Hopefully, most remaining issues 
can be found before 4.7.0 is released.
-------------- next part --------------
A non-text attachment was scrubbed...
Name: constructor-v4a.diff
Type: text/x-patch
Size: 66452 bytes
Desc: not available
URL: <http://gcc.gnu.org/pipermail/fortran/attachments/20111031/13273d0c/attachment.bin>