[Bug fortran/54384] New: gfortran memory leaks

[Bug fortran/54384] New: gfortran memory leaks

[burnus at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54384

             Bug #: 54384
           Summary: gfortran memory leaks
    Classification: Unclassified
           Product: gcc
           Version: 4.8.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: fortran
        AssignedTo: unassigned@gcc.gnu.org
        ReportedBy: burnus@gcc.gnu.org


Running valgrind on gfortran (or rather f951) -fsyntax-only shows the following
leaks.

(This PR assumed that my patch
http://gcc.gnu.org/ml/fortran/2012-08/msg00180.html has been applied plus a
follow up for module.c's mio_symbol which removes the "sym->refs++".)


With Polyhedron 2005's channel.f90 (and rnflow.f90):
  by 0x606B33: gfc_get_array_ss(gfc_ss*, gfc_expr*, int, gfc_ss_type)
(trans-array.c:561)
  by 0x612F00: gfc_walk_expr(gfc_expr*) (trans-array.c:8787)
  by 0x639F98: gfc_trans_arrayfunc_assign(gfc_expr*, gfc_expr*)
(trans-expr.c:6749)
  by 0x63B911: gfc_trans_assignment(gfc_expr*, gfc_expr*, bool, bool)
(trans-expr.c:7438)


With Polyhedron 2005's mdbx.f90:
  by 0x57D7BF: gfc_get_expr() (expr.c:49)
  by 0x57E123: gfc_copy_expr(gfc_expr*) (expr.c:272)
  by 0x5C63EA: gfc_match_rvalue(gfc_expr**) (primary.c:2788)
  by 0x56D3A2: match_data_constant(gfc_expr**) (decl.c:354)


With Polyhedron 2005's nf.f90:
  by 0x57E1807: __gmpz_init (in /usr/lib64/libgmp.so.10.0.5)
  by 0x57D9EB: gfc_get_constant_expr(bt, int, locus*) (expr.c:161)
  by 0x57DB17: gfc_get_int_expr(int, locus*, int) (expr.c:217)
  by 0x5F592C: _ZL24check_section_vs_sectionP13gfc_array_refS0_i.part.4
(dependency.c:1213)
  by 0x5F6C47: gfc_dep_resolver(gfc_ref*, gfc_ref*, gfc_reverse*)
(dependency.c:1155)
  by 0x60C585: gfc_conv_resolve_dependencies(gfc_loopinfo*, gfc_ss*, gfc_ss*)
(trans-array.c:4309)
  by 0x63A891: gfc_trans_assignment_1(gfc_expr*, gfc_expr*, bool, bool)
(trans-expr.c:7238)


With Polyhedron 2005's capacita.f90:  Several. I think the only new one is:
  by 0x5EF33E: gfc_get_namespace(gfc_namespace*, int) (symbol.c:2325)
  by 0x5F0F09: gfc_copy_formal_args_intr(gfc_symbol*, gfc_intrinsic_sym*)
(symbol.c:4161)
  by 0x63C50C: gfc_get_symbol_for_expr(gfc_expr*) (trans-intrinsic.c:2334)
  by 0x64A0EB: gfc_conv_intrinsic_function(gfc_se*, gfc_expr*)
(trans-intrinsic.c:4525)


With Polyhedron 2005's capacita.f90 (and some more):
  by 0x57D9EB: gfc_get_constant_expr(bt, int, locus*) (expr.c:161)
  by 0x57DB17: gfc_get_int_expr(int, locus*, int) (expr.c:217)
  by 0x55C016: gfc_simplify_iterator_var(gfc_expr*) (array.c:1445)
  by 0x5826E4: gfc_simplify_expr(gfc_expr*, int) (expr.c:1894)
  by 0x56A394: gfc_assign_data_value(gfc_expr*, gfc_expr*, __mpz_struct*,
__mpz_struct (*) [1]) (data.c:61)



With gfortran.dg/proc_ptr_result_1.f90:
  by 0x5EF6E7: gfc_new_symbol(char const*, gfc_namespace*) (symbol.c:2569)
  by 0x5716DA: gfc_match_procedure() (decl.c:4916)

---- and ----

  by 0x5EF6E7: gfc_new_symbol(char const*, gfc_namespace*) (symbol.c:2569)
  by 0x5EFA62: gfc_get_sym_tree(char const*, gfc_namespace*, gfc_symtree**,
bool) (symbol.c:2753)
  by 0x56B76C: find_special(char const*, gfc_symbol**, bool) (decl.c:802)
  by 0x574F28: attr_decl() (decl.c:6174)

---- and ----

  by 0x5EF33E: gfc_get_namespace(gfc_namespace*, int) (symbol.c:2325)
  by 0x5F0D3C: gfc_copy_formal_args(gfc_symbol*, gfc_symbol*, ifsrc)
(symbol.c:4104)
  by 0x5F0E30: gfc_copy_formal_args(gfc_symbol*, gfc_symbol*, ifsrc)
(symbol.c:4117)
  by 0x5D7A5E: resolve_procedure_interface(gfc_symbol*) (resolve.c:228)
  by 0x5CC420: resolve_symbol(gfc_symbol*) (resolve.c:12581)

---- and ----

  by 0x5EF33E: gfc_get_namespace(gfc_namespace*, int) (symbol.c:2325)
  by 0x5B4040: load_needed(pointer_info*) (module.c:4375)
  by 0x5B4DCB: read_module() (module.c:4773)

---- and ----

  the gfc_match_rvalue issue of mdbx.f90.



Also leaking, but I lost the overview whether those are already covered:
doduc.f90. (I also haven't tested all of Polyhedron.)

[Bug fortran/54384] gfortran memory leaks

[burnus at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54384

--- Comment #1 from Tobias Burnus <burnus at gcc dot gnu.org> 2012-08-27 14:42:54 UTC ---
Author: burnus
Date: Mon Aug 27 14:42:50 2012
New Revision: 190713

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=190713
Log:
2012-08-27  Tobias Burnus  <burnus@net-b.de>

        PR fortran/54384
        * trans-expr.c (gfc_trans_arrayfunc_assign): Free se.ss
        and loop.


Modified:
    trunk/gcc/fortran/ChangeLog
    trunk/gcc/fortran/trans-expr.c

[Bug fortran/54384] gfortran memory leaks

[burnus at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54384

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |burnus at gcc dot gnu.org

--- Comment #2 from Tobias Burnus <burnus at gcc dot gnu.org> 2012-08-27 14:57:21 UTC ---
The patch of comment 0 fixed the
gfc_trans_assignment->gfc_trans_arrayfunc_assign issue of channel.f90 and
rnflow.f90. The other issues remain.

[Bug fortran/54384] gfortran memory leaks

[burnus at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54384

--- Comment #3 from Tobias Burnus <burnus at gcc dot gnu.org> 2012-08-27 18:48:03 UTC ---
Author: burnus
Date: Mon Aug 27 18:47:57 2012
New Revision: 190728

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=190728
Log:
2012-08-27  Tobias Burnus  <burnus@net-b.de>

        PR fortran/54384
        * dependency.c (check_section_vs_section): Use gfc_free_expr
        instead of free.
        * trans-intrinsic.c (conv_generic_with_optional_char_arg): Use
        gfc_free_symbol instead of free.


Modified:
    trunk/gcc/fortran/ChangeLog
    trunk/gcc/fortran/dependency.c
    trunk/gcc/fortran/trans-intrinsic.c

[Bug fortran/54384] gfortran memory leaks

[burnus at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54384

--- Comment #4 from Tobias Burnus <burnus at gcc dot gnu.org> 2012-08-27 18:52:29 UTC ---
(In reply to comment #0)
> With Polyhedron 2005's nf.f90:
> With Polyhedron 2005's capacita.f90:

Those are fixed by comment 3


> With Polyhedron 2005's capacita.f90 (and some more):

I think that's aermod.f90 not capacita.f90 - the latter has no leaks any more.


> With Polyhedron 2005's mdbx.f90:
>   by 0x57D7BF: gfc_get_expr() (expr.c:49)
>   by 0x57E123: gfc_copy_expr(gfc_expr*) (expr.c:272)
>   by 0x5C63EA: gfc_match_rvalue(gfc_expr**) (primary.c:2788)
>   by 0x56D3A2: match_data_constant(gfc_expr**) (decl.c:354)

Reduced test case:

      DIMENSION  cmul(3,7)
      PARAMETER (HALF=0.5D0,ONE=1.D0,TWO=2.D0,FOUR=4.D0,TWELVE=12.D0)
      PARAMETER (SQRT2=1.41421356237310D0,SQRT3=1.73205080756888D0,     &
     &           SQRT8=TWO*SQRT2,SQ8BY3=SQRT8/SQRT3,SQ4BY3=TWO/SQRT3,   &
     &           SQ16B3=FOUR/SQRT3)
      DATA cmul/SQRT2 , SQRT2 , SQRT2 , ONE , ONE , SQRT8 , ONE ,       &
     &     SQRT2 , TWO , ONE , SQRT3 , SQ8BY3 , ONE , SQRT3 , SQ8BY3 ,  &
     &     ONE , ONE , SQRT8 , SQ4BY3 , SQ4BY3 , SQ16B3/
      END

Not fixing this issue, but I think still required:

--- a/gcc/fortran/data.c
+++ b/gcc/fortran/data.c
@@ -142,2 +142,4 @@ create_character_initializer (gfc_expr *init, gfc_typespec
*ts,
       gfc_extract_int (end_expr, &end);
+      gfc_free_expr (start_expr);
+      gfc_free_expr (end_expr);
     }


> Also leaking, but I lost the overview whether those are already covered:
> doduc.f90.

Others still not tested, but doduc.f90 fails with:

  by 0x63624D: _ZL22gfc_conv_function_exprP6gfc_seP8gfc_expr.part.22
(trans-expr.c:4999)
  by 0x63951B: gfc_conv_expr_op(gfc_se*, gfc_expr*) (trans-expr.c:2158)
  by 0x63951B: gfc_conv_expr_op(gfc_se*, gfc_expr*) (trans-expr.c:2158)
  by 0x638553: gfc_conv_expr_val(gfc_se*, gfc_expr*) (trans-expr.c:5851)
  by 0x638620: gfc_conv_unary_op(tree_code, gfc_se*, gfc_expr*)
(trans-expr.c:1535)
  by 0x639AFF: gfc_conv_expr_op(gfc_se*, gfc_expr*) (trans-expr.c:2031)
  by 0x6394F6: gfc_conv_expr_op(gfc_se*, gfc_expr*) (trans-expr.c:2153)
  by 0x63C163: gfc_trans_assignment_1(gfc_expr*, gfc_expr*, bool, bool)
(trans-expr.c:7278)

[Bug fortran/54384] gfortran memory leaks

[burnus at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54384

--- Comment #5 from Tobias Burnus <burnus at gcc dot gnu.org> 2012-08-27 20:51:57 UTC ---
Author: burnus
Date: Mon Aug 27 20:51:52 2012
New Revision: 190731

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=190731
Log:
2012-08-27  Tobias Burnus  <burnus@net-b.de>

        PR fortran/54384
        * resolve.c (gfc_resolve_character_operator): Free temporary
        variables.
        * trans-expr.c (gfc_conv_statement_function): Ditto.


Modified:
    trunk/gcc/fortran/ChangeLog
    trunk/gcc/fortran/resolve.c
    trunk/gcc/fortran/trans-expr.c

[Bug fortran/54384] gfortran memory leaks

[burnus at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54384

--- Comment #6 from Tobias Burnus <burnus at gcc dot gnu.org> 2012-08-27 20:54:53 UTC ---
(In reply to comment #4)
> Others still not tested, but doduc.f90 fails with:

This one (and hence all of doduc.f90) - and some of the failures in aermod.f90
are fixed with comment 5.


TODO:

* match_data_constant (mbdx.f90, comment 5)

* gfortran.dg/proc_ptr_result_1.f90 (comment 0)

* aermod.f90:

  by 0x5F0EC7: gfc_new_symbol(char const*, gfc_namespace*) (symbol.c:2569)
  by 0x5F1242: gfc_get_sym_tree(char const*, gfc_namespace*, gfc_symtree**,
bool) (symbol.c:2753)
  by 0x5F12C3: gfc_get_symbol(char const*, gfc_namespace*, gfc_symbol**)
(symbol.c:2807)
  by 0x57184A: gfc_match_formal_arglist(gfc_symbol*, int, int) (decl.c:4462)
  by 0x573E65: gfc_match_subroutine() (decl.c:5655)
-- and --
  by 0x5F0B1E: gfc_get_namespace(gfc_namespace*, int) (symbol.c:2325)
  by 0x5C064E: parse_spec(gfc_statement) (parse.c:2390)
  by 0x5C2318: parse_progunit(gfc_statement) (parse.c:4131)
  by 0x5C3243: gfc_parse_file() (parse.c:4579)
-- and --
  by 0x57F00F: gfc_get_expr() (expr.c:49)
  by 0x57F973: gfc_copy_expr(gfc_expr*) (expr.c:272)
  by 0x56BBD4: gfc_assign_data_value(gfc_expr*, gfc_expr*, __mpz_struct*,
__mpz_struct (*) [1]) (data.c:60)
  by 0x5D86DB: traverse_data_var(gfc_data_variable*, locus*) (resolve.c:13341)
  by 0x5D8C0D: traverse_data_var(gfc_data_variable*, locus*) (resolve.c:13426)
  by 0x5D9E83: resolve_types(gfc_namespace*) (resolve.c:13525)
-- and --
  some more


* gfortran.dg/g77/cpp6.f:
  by 0x5DE174: preprocessor_line(unsigned int*) (scanner.c:1803)
  by 0x5DE6EA: load_file(char const*, char const*, bool) (scanner.c:2038)
  by 0x5E0422: gfc_new_file() (scanner.c:2106)

[Bug fortran/54384] gfortran memory leaks

[burnus at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54384

--- Comment #7 from Tobias Burnus <burnus at gcc dot gnu.org> 2012-08-28 05:35:22 UTC ---
Author: burnus
Date: Tue Aug 28 05:35:15 2012
New Revision: 190738

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=190738
Log:
2012-08-28  Tobias Burnus  <burnus@net-b.de>

        PR fortran/54384
        * symbol.c (gfc_copy_formal_args): Set also sym->formal_ns.


Modified:
    trunk/gcc/fortran/ChangeLog
    trunk/gcc/fortran/symbol.c

[Bug fortran/54384] gfortran memory leaks

[burnus at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54384

--- Comment #8 from Tobias Burnus <burnus at gcc dot gnu.org> 2012-08-28 05:37:37 UTC ---
(In reply to comment #0)
> With gfortran.dg/proc_ptr_result_1.f90:
>   by 0x5F0D3C: gfc_copy_formal_args(gfc_symbol*, gfc_symbol*, ifsrc)

That issue is now fixed by comment 7 - the others remain.

[Bug fortran/54384] gfortran memory leaks

[burnus at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54384

--- Comment #9 from Tobias Burnus <burnus at gcc dot gnu.org> 2012-08-28 14:49:19 UTC ---
Author: burnus
Date: Tue Aug 28 14:49:15 2012
New Revision: 190751

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=190751
Log:
2012-08-28  Tobias Burnus  <burnus@net-b.de>

        PR fortran/54384
        * decl.c (match_data_constant): Add missing gfc_free_expr.
        (top_val_list): Remove always-true condition.
        * data.c (get_array_index, create_character_initializer):
        Free temporary expressions.
        (gfc_assign_data_value): Free expression when aborting.


Modified:
    trunk/gcc/fortran/ChangeLog
    trunk/gcc/fortran/data.c
    trunk/gcc/fortran/decl.c

[Bug fortran/54384] gfortran memory leaks

[burnus at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54384

--- Comment #10 from Tobias Burnus <burnus at gcc dot gnu.org> 2012-08-28 14:51:31 UTC ---
(In reply to comment #6)
> * match_data_constant (mbdx.f90, comment 5)
> * aermod.f90:
>   by 0x5D8C0D: traverse_data_var(gfc_data_variable*, locus*) (resolve.c:13426)

These two are now fixed. (comment 9)


TODO:

* aermod.f90 (cf. comment 6)
  - parse_spec -> gfc_match_subroutine -> gfc_match_formal_arglist
  - parse_spec -> gfc_get_namespace  (also: gas_dyn.f90 and test_fpu.f90)
  those are the only gfortran valgrind-reported leaks left for Polyhedron 2005.

* gfortran.dg/g77/cpp6.f's preprocessor_line, cf. comment 6

* gfortran.dg/proc_ptr_result_1.f90 (comment 0) with
  - ...-> match_level_3 -> ... -> gfc_match_rvalue -> gfc_new_symbol
  - read_module -> load_needed -> gfc_get_namespace
  - gfc_match_procedure -> gfc_new_symbol
  - gfc_match_intrinsic -> gfc_new_symbol


[Plus more valgrind runs to find more leaks.]

[Bug fortran/54384] gfortran memory leaks

[dominiq at lps dot ens.fr]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54384

Dominique d'Humieres <dominiq at lps dot ens.fr> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
   Last reconfirmed|                            |2013-06-29
     Ever confirmed|0                           |1

--- Comment #11 from Dominique d'Humieres <dominiq at lps dot ens.fr> ---
Considering the number of commits, this PR can probably marked as confirmed.