[Bug c++/40058] New: C++ FE generates struct assignments with mismatched sizes

[Bug c++/40058] New: C++ FE generates struct assignments with mismatched sizes

[rguenth at gcc dot gnu dot org]

struct Base {
 Base();
 int i;
 char c;
};

template <typename BaseT>
struct actor : public BaseT {
 actor(BaseT const& base);
 char c;
};

template <typename BaseT>
actor<BaseT>::actor(BaseT const& base)
    : BaseT(base) {}

actor<Base>
foo(const Base c)
{
 return actor<Base>(c);
}

with -O2 we inline the constructor and get

actor<Base> foo(Base) D.2099 (const struct Base c)
{
 struct actor D.2133;
 struct actor D.2134;

<bb 2>:
 D.2133.D.2111 = c;
 D.2134 = D.2133;
 return D.2134;

}

The assignment

 D.2133.D.2111 = c;

has on the LHS a size of 40 (from the FIELD_DECL) and on the
RHS a size of 64, from the decl.  In the un-inlined constructor we had
the bogus

actor<BaseT>::actor(const BaseT&) [with BaseT = Base] D.2105 (struct
actor * const this, const struct Base & base)
{
<bb 2>:
 this_1(D)->D.2111 = *base_2(D);
 return;

}

with the same issue (size 40 on the lhs, size 64 on the rhs).

"Size" in term of what the FIELD_DECL DECL_SIZE says for the component-refs,
which is according to people what is relevant - not the type size.  Likewise
for the RHS the DECL_SIZE (which is equal to TYPE_SIZE).

For less contrived examples the FE generates a memcpy to carefully _not_
overwrite padding that may be used in tail packing.

Eventually from the above you can construct a wrong-code testcase if expansion
chooses the wrong size for the expansion of the assignment.


-- 
           Summary: C++ FE generates struct assignments with mismatched
                    sizes
           Product: gcc
           Version: 4.5.0
            Status: UNCONFIRMED
          Keywords: wrong-code
          Severity: normal
          Priority: P3
         Component: c++
        AssignedTo: unassigned at gcc dot gnu dot org
        ReportedBy: rguenth at gcc dot gnu dot org


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=40058

[Bug c++/40058] C++ FE generates struct assignments with mismatched sizes

[jason at gcc dot gnu dot org]

------- Comment #1 from jason at gcc dot gnu dot org  2010-02-10 05:46 -------
Your comment to the patch that removed LANG_HOOKS_EXPR_SIZE suggests that this
isn't a problem anymore:

http://gcc.gnu.org/ml/gcc-patches/2009-08/msg01593.html


-- 

jason at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |jason at gcc dot gnu dot org


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=40058

[Bug c++/40058] C++ FE generates struct assignments with mismatched sizes

[rguenth at gcc dot gnu dot org]

------- Comment #2 from rguenth at gcc dot gnu dot org  2010-02-10 09:42 -------
We still have these constructs and expand still properly deals with them
(it uses the size of the LHS).

I'm not sure what's the best thing to do about this bug.  Eventually better
documenting what sizes in an assignment are really relevant.

The FE now indeed emits memcpy whenever there is not a DECL or FIELD_DECL
on the LHS that specifies sizes, like for

void bar (Base *p, Base *q)
{
  *p = *q;
}

;; Function void bar(Base*, Base*) (null)
;; enabled by -tree-original

<<cleanup_point <<< Unknown tree: expr_stmt
  (void) (struct Base *) __builtin_memcpy (NON_LVALUE_EXPR <p>, (const struct
Base &) (const struct Base *) NON_LVALUE_EXPR <q>, 5) >>>
>>;


So the langhook was no longer necessary.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=40058

[Bug c++/40058] C++ FE generates struct assignments with mismatched sizes

[jamborm at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=40058

--- Comment #3 from Martin Jambor <jamborm at gcc dot gnu.org> 2012-03-21 10:05:24 UTC ---
If this ever gets fixed, we can remove a test in SRA as in:

http://gcc.gnu.org/ml/gcc-patches/2012-03/msg01356.html