[Bug fortran/51175] Memory Leak with Transfer intrinsic

[burnus at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=51175

Tobias Burnus <burnus at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Keywords|                            |wrong-code
             Status|UNCONFIRMED                 |NEW
   Last reconfirmed|                            |2011-11-16
                 CC|                            |burnus at gcc dot gnu.org
     Ever Confirmed|0                           |1

--- Comment #1 from Tobias Burnus <burnus at gcc dot gnu.org> 2011-11-16 17:15:31 UTC ---
(In reply to comment #0)
> I am using gfortran 4.6.2 on MacOSX 10.7. 
> It appears that this code involving transfer generates a memory leak

Indeed, valgrind shows:

==3168== 20 bytes in 1 blocks are definitely lost in loss record 1 of 1
==3168==    at 0x4A0610C: malloc (vg_replace_malloc.c:195)
==3168==    by 0x4009BF: __leakmodule_MOD_leaking (test.f90:14)
==3168==    by 0x400889: __leakmodule_MOD_foo (test.f90:26)
==3168==    by 0x400A5B: MAIN__ (test.f90:34)
==3168==    by 0x400A91: main (test.f90:32)


>       print *, outstring

Use:
  print *, outstring(1:3)
otherwise you access uninitialized memory.


> According to the produced assembler, malloc gets called inside the Leaking
> routine, but the asm of the caller never issues a free of that allocated
> memory.

Easier than studying the assembler: Use -fdump-tree-original  (or
-fdump-tree-optimized) which creates a file of the internal representation with
C-like syntax. [Note: The dump does not contain all details.]