[Bug fortran/48112] [4.6/4.7 Regression] generic interface to external function in module

[burnus at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=48112

--- Comment #4 from Tobias Burnus <burnus at gcc dot gnu.org> 2011-04-26 10:18:23 UTC ---
Rev. 159475 works, Rev. 159476 fails, which is the following commit:

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=159476
Log:
2010-05-17  Janus Weil  <janus@gcc.gnu.org>

    PR fortran/44044
    * resolve.c (resolve_fl_var_and_proc): Move error messages here from ...
    (resolve_fl_variable_derived): ... this place.
    (resolve_symbol): Make sure function symbols (and their result
    variables) are not resolved twice.


The crucial part of the patch is:

+  /* Avoid double resolution of function result symbols.  */
+  if ((sym->result || sym->attr.result) && (sym->ns != gfc_current_ns))
+    return;

 * * *

If one places a break point, one sees first:

(gdb) p sym->name
$5 = 0x2aaaaab40ea8 "test1"
(gdb) p sym->ns->sym_root->name
$8 = 0x2aaaaab40ea8 "test1"
(gdb) p ((sym->result || sym->attr.result) && (sym->ns != gfc_current_ns))
$10 = 0

and then next:

(gdb) p sym->name
$14 = 0x2aaaaab40ea0 "test"
(gdb) p sym->ns->sym_root->name
$15 = 0x2aaaaab40ea0 "test"
(gdb) p gfc_current_ns->sym_root->name
$16 = 0x2aaaaab40ea8 "test1"
And hence:
(gdb) p ((sym->result || sym->attr.result) && (sym->ns != gfc_current_ns))
$17 = 1

and finally:

(gdb) p sym->name
$26 = 0x2aaaaab40ea8 "test1"
(gdb) p sym->ns->sym_root->name
$27 = 0x2aaaaab40ea8 "test1"
(gdb) p gfc_current_ns->sym_root->name
$28 = 0x2aaaaab40ea0 "test"
(gdb) p ((sym->result || sym->attr.result) && (sym->ns != gfc_current_ns))
$25 = 1


Seemingly, the ICE occurs unless both resolutions happen.


(I wonder whether one could use {sym->ns,gfc_current_namespace}->resolved to
decide whether one should continue or not; currently, one finds the values "-1"
and "0".)