New ICE with gfortran 5.0 v221644

[Juergen Reuter]

Wolfgang, this is mega urgent to find a workaround!!!!

libtool: compile: gfortran -I../basics -I../utilities -I../testing -I../system -I../../vamp/src -fopenmp -g -O2 -c selectors.f90 -fno-common -o .libs/selectors.o selectors.f90:94:0:

source = pack ([(i, i = 1, size (weight))], mask))

1

internal compiler error: in gfc_conv_procedure_call, at fortran/trans-expr.c:5608

[Juergen Reuter]

We have been working around this one, but I expect this to be fixed anyhow before 5.0 comes out (keep fingers crossed). However, there is a remaining segfault connected to the smtest_13:

| Events: closing raw file 'smtest_13_1.evx'
whizard(4764) malloc: *** error for object 0x3134353933360006: pointer being freed was not allocated
*** set a breakpoint in malloc_error_break to debug

Program received signal SIGABRT: Process abort signal.

Backtrace for this error:
#0  0x102b2e202
#1  0x102b2d520
#2  0x7fff895d5cf9

where the backtrace can be (partially) accessed:

0x00007fff8f89d82a in __kill ()
(gdb) bt
#0  0x00007fff8f89d82a in __kill ()
#1  0x00007fff89574a9c in abort ()
#2  0x00007fff895d384c in free ()
#3  0x0000000100653dc3 in __eval_trees_MOD_eval_tree_final (expr={_data = 0x102b85650, _vptr = 0x100caad40}) at eval_trees.f90:6650

This is unrelated to the stuff in #711 as it appears already in r6751 before the (try to) refactor interactions.

[Juergen Reuter]

I don't know if that helps: but if I print the eval_tree just before the finalization, the problem doesn't occur:

[...]
|  |  |  o sw =>  4.714339240339E-01
|  |  o mW =>  8.041900000000E+01
|  o v =>  2.462184581018E+02
 en_binary
 en_binary
 en_binary
 Evaluation tree:
o [/] =  1.325049458125E+02
|  o [const] =  1.256637061436E+01
|  o [**] =  9.483699296887E-02
|  |  o ee =>  3.079561542961E-01
|  |  o [const] = 2
 en_binary
 en_binary
| There were no errors and    5 warning(s).
| WHIZARD run finished.
|=============================================================================|

[Juergen Reuter]

Yep, please blame me. The problem was the mismatch in sizes between the weight array and the mask array in the selector_int 'fix' for the gfortran 5.0 issue. Happily closing this one.