[Bug translator/14434] New: dwflpp sometimes caches incomplete class_type

[Bug translator/14434] New: dwflpp sometimes caches incomplete class_type

[mjw at redhat dot com]

http://sourceware.org/bugzilla/show_bug.cgi?id=14434

             Bug #: 14434
           Summary: dwflpp sometimes caches incomplete class_type
           Product: systemtap
           Version: unspecified
            Status: NEW
          Severity: normal
          Priority: P2
         Component: translator
        AssignedTo: systemtap@sourceware.org
        ReportedBy: mjw@redhat.com
    Classification: Unclassified


Take the following program:

::::::::::::::
header.hxx
::::::::::::::
#include <malloc.h>

class Heap
{
  private:
    char *_memory;
    size_t _size;
  public:
    Heap();
    void* allocate  (size_t size);
    static size_t header_size();
};
::::::::::::::
heap.cxx
::::::::::::::
#include "header.hxx"

size_t
Heap::header_size ()
{
  return 42;
}

Heap::Heap()
{
  _size = header_size () + 32;
  _memory = (char *) malloc (_size);
}

void*
Heap::allocate(size_t size)
{
  _size += size;
  return _memory + header_size();
}
::::::::::::::
main.cxx
::::::::::::::
#include "header.hxx"

int size (int resize)
{
  return (int)Heap::header_size() - resize;
}

int
main (int argc, char **argv)
{
  return size (argc);
}

Compile it with g++ 4.7.1:

$ g++ -g -O2 -c main.cxx
$ g++ -g -O2 -c heap.cxx
$ g++ -g main.o heap.o -o prog

Now try inspecting it with stap a couple of times and notice it occasionally
fails. In this example we are interested in the offset of one of the field
members of the Heap class and use @cast on address zero to get it:

$ stap -p2 -v -e 'probe process("./prog").function("main") { printf("_size
member offset: %d\n", &@cast(0, "Heap")->_size); }' > /dev/null
Pass 1: parsed user script and 96 library script(s) using
212776virt/34756res/3024shr/32128data kb, in 110usr/10sys/127real ms.
Pass 2: analyzed script: 1 probe(s), 1 function(s), 0 embed(s), 0 global(s)
using 215824virt/36464res/3588shr/33056data kb, in 10usr/0sys/7real ms.
$ stap -p2 -v -e 'probe process("./prog").function("main") { printf("_size
member offset: %d\n", &@cast(0, "Heap")->_size); }' > /dev/null
Pass 1: parsed user script and 96 library script(s) using
212776virt/34756res/3024shr/32128data kb, in 110usr/10sys/129real ms.
Pass 2: analyzed script: 1 probe(s), 1 function(s), 0 embed(s), 0 global(s)
using 215824virt/36464res/3588shr/33056data kb, in 10usr/0sys/7real ms.
$ stap -p2 -v -e 'probe process("./prog").function("main") { printf("_size
member offset: %d\n", &@cast(0, "Heap")->_size); }' > /dev/null
Pass 1: parsed user script and 96 library script(s) using
212776virt/34756res/3024shr/32128data kb, in 120usr/10sys/128real ms.
semantic error: unable to find member '_size' for class Heap: operator '->' at
<input>:1:97
        source: probe process("./prog").function("main") { printf("_size member
offset: %d\n", &@cast(0, "Heap")->_size); }
                                                                               
                                ^

Pass 2: analyzed script: 1 probe(s), 0 function(s), 0 embed(s), 0 global(s)
using 213716virt/36340res/3476shr/33052data kb, in 10usr/0sys/7real ms.
Pass 2: analysis failed.  Try again with another '--vp 01' option.

Isn't that weird! It worked 2 out of 3 times, but then it suddenly failed.

-- 
Configure bugmail: http://sourceware.org/bugzilla/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are the assignee for the bug.

[Bug translator/14434] dwflpp sometimes caches incomplete class_type

[mjw at redhat dot com]

http://sourceware.org/bugzilla/show_bug.cgi?id=14434

--- Comment #1 from Mark Wielaard <mjw at redhat dot com> 2012-08-05 18:51:29 UTC ---
The reason it sometimes fails is that dwflpp::global_alias_caching_callback
puts types in an unordered_map<std::string, Dwarf_Die> (cu_type_cache_t) and
depending on the order this map is filled in it might pick the "wrong"
class_type for Heap.

It can pick the wrong one since there are two in this example:

$ eu-readelf --debug-dump=info prog

[...]
 [     b]  compile_unit
           producer             (strp) "GNU C++ 4.7.1 20120629 (Red Hat
4.7.1-1) -fpreprocessed -mtune=generic -march=x86-64 -g -O2"
           language             (data1) C++ (4)
           name                 (strp) "main.cxx"
           comp_dir             (strp) "/home/mark/src/tests/cxx-decl"
           ranges               (sec_offset) range list [     0]
           low_pc               (addr) 000000000000000000
           stmt_list            (sec_offset) 0
[...]
 [   289]    class_type
             name                 (strp) "Heap"
             byte_size            (data1) 16
             decl_file            (data1) 6
             decl_line            (data1) 3
             sibling              (ref4) [   2a6]
 [   295]      subprogram
               external             (flag_present) Yes
               name                 (strp) "header_size"
               decl_file            (data1) 6
               decl_line            (data1) 11
               linkage_name         (strp) "_ZN4Heap11header_sizeEv"
               type                 (ref4) [    30]
               accessibility        (data1) public (1)
               declaration          (flag_present) Yes
 [   38e]  compile_unit
           producer             (strp) "GNU C++ 4.7.1 20120629 (Red Hat
4.7.1-1) -fpreprocessed -mtune=generic -march=x86-64 -g -O2"
           language             (data1) C++ (4)
           name                 (strp) "heap.cxx"
           comp_dir             (strp) "/home/mark/src/tests/cxx-decl"
           low_pc               (addr) 0x00000000004005b0
<_ZN4Heap11header_sizeEv>
           high_pc              (addr) 0x00000000004005ec
           stmt_list            (sec_offset) 220
[...]
 [   610]    class_type
             name                 (strp) "Heap"
             byte_size            (data1) 16
             decl_file            (data1) 5
             decl_line            (data1) 3
             sibling              (ref4) [   67e]
 [   61c]      member
               name                 (strp) "_memory"
               decl_file            (data1) 5
               decl_line            (data1) 6
               type                 (ref4) [   412]
               data_member_location (data1) 0
 [   628]      member
               name                 (strp) "_size"
               decl_file            (data1) 5
               decl_line            (data1) 7
               type                 (ref4) [   3b7]
               data_member_location (data1) 8
 [   634]      subprogram
               external             (flag_present) Yes
               name                 (strp) "Heap"
               decl_file            (data1) 5
               decl_line            (data1) 9
               accessibility        (data1) public (1)
               declaration          (flag_present) Yes
               object_pointer       (ref4) [   644]
               sibling              (ref4) [   64a]
 [   644]        formal_parameter
                 type                 (ref4) [   67e]
                 artificial           (flag_present) Yes
 [   64a]      subprogram
               external             (flag_present) Yes
               name                 (strp) "allocate"
               decl_file            (data1) 5
               decl_line            (data1) 10
               linkage_name         (strp) "_ZN4Heap8allocateEm"
               type                 (ref4) [   410]
               accessibility        (data1) public (1)
               declaration          (flag_present) Yes
               object_pointer       (ref4) [   662]
               sibling              (ref4) [   66d]
 [   662]        formal_parameter
                 type                 (ref4) [   67e]
                 artificial           (flag_present) Yes
 [   667]        formal_parameter
                 type                 (ref4) [   3b7]
 [   66d]      subprogram
               external             (flag_present) Yes
               name                 (strp) "header_size"
               decl_file            (data1) 5
               decl_line            (data1) 11
               linkage_name         (strp) "_ZN4Heap11header_sizeEv"
               type                 (ref4) [   3b7]
               accessibility        (data1) public (1)
               declaration          (flag_present) Yes

Note how the first one in the main.cxx compile_unit is incomplete.
It only contains the header_size subprogram member (since that is the only
member used in main.cxx).

By just looking at the class_type DIE we cannot tell which one we need though,
since both [   289] and [   610] look identical, both have a byte_size
attribute and neither is marked as declaration.

I think this is a gcc DWARF generation bug:
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54181
" partial DW_TAG_class_type generated with DW_AT_byte_size and without
DW_AT_declaration"

-- 
Configure bugmail: http://sourceware.org/bugzilla/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are the assignee for the bug.

[Bug translator/14434] dwflpp sometimes caches incomplete class_type

[mjw at redhat dot com]

http://sourceware.org/bugzilla/show_bug.cgi?id=14434

--- Comment #2 from Mark Wielaard <mjw at redhat dot com> 2012-08-05 18:55:45 UTC ---
Proposed patch/workaround:

diff --git a/dwflpp.cxx b/dwflpp.cxx
index 3fb50b2..ea93469 100644
--- a/dwflpp.cxx
+++ b/dwflpp.cxx
@@ -791,6 +791,32 @@ cache_type_prefix(Dwarf_Die* type)
   return "";
 }

+/* GCC might generate a struct/class without DW_AT_declaration,
+   but that only contains members which have DW_AT_declaration
+   set.  We aren't interested in those.  PR14434 (GCC bug #54181).  */
+static bool
+has_only_decl_members (Dwarf_Die *die)
+{
+  Dwarf_Die child;
+  if (dwarf_child(die, &child) != 0)
+    return false; /* no members */
+
+  do
+    {
+      if (! dwarf_hasattr(&child, DW_AT_declaration))
+    return false; /* real member found.  */
+      int tag = dwarf_tag(&child);
+      if ((tag == DW_TAG_namespace
+           || tag == DW_TAG_structure_type
+           || tag == DW_TAG_class_type)
+          && ! has_only_decl_members (&child))
+    return false; /* real grand child member found.  */
+    }
+  while (dwarf_siblingof(&child, &child) == 0);
+
+  return true; /* Tried all children and grandchildren. */
+}
+
 int
 dwflpp::global_alias_caching_callback(Dwarf_Die *die, bool has_inner_types,
                                       const string& prefix, void *arg)
@@ -798,7 +824,8 @@ dwflpp::global_alias_caching_callback(Dwarf_Die *die, bool
has_inner_types,
   cu_type_cache_t *cache = static_cast<cu_type_cache_t*>(arg);
   const char *name = dwarf_diename(die);

-  if (!name || dwarf_hasattr(die, DW_AT_declaration))
+  if (!name || dwarf_hasattr(die, DW_AT_declaration)
+      || has_only_decl_members(die))
     return DWARF_CB_OK;

   int tag = dwarf_tag(die);

-- 
Configure bugmail: http://sourceware.org/bugzilla/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are the assignee for the bug.

[Bug translator/14434] dwflpp sometimes caches incomplete class_type

[mjw at redhat dot com]

http://sourceware.org/bugzilla/show_bug.cgi?id=14434

Mark Wielaard <mjw at redhat dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|NEW                         |RESOLVED
         Resolution|                            |FIXED

--- Comment #3 from Mark Wielaard <mjw at redhat dot com> 2012-08-06 22:39:34 UTC ---
commit 87eeec9407f8afa5af1037600272eebf0561e1d1
Author: Mark Wielaard <mjw@redhat.com>
Date:   Mon Aug 6 20:31:20 2012 +0200

    PR14434 Filter out partial structs/classes.

    GCC might generate a struct/class without DW_AT_declaration,
    but that only contains members which have DW_AT_declaration
    set.  We aren't interested in those.  PR14434 (GCC bug #54181).
    Filter them out with a new function has_only_decl_members in dwflpp.cxx.
    Add new testcase testsuite/systemtap.base/partial-class-type.exp.

-- 
Configure bugmail: http://sourceware.org/bugzilla/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are the assignee for the bug.