bpf_dsl: fix argument-dependent lookup of BoolExpr operators

bpf_dsl: fix argument-dependent lookup of BoolExpr operators

Committed: https://crrev.com/b3038b5483c9601e64f05b3eb23e1a1f5b11e5ce
Cr-Commit-Position: refs/heads/master@{#294127}

[mdempsky, 2014-09-06 01:33:21]

mdempsky@chromium.org changed reviewers:
+ jln@chromium.org, mseaborn@chromium.org

[Mark Seaborn, 2014-09-09 20:29:18]

I don't know the BPF DSL code very well.  What does this fix?  Could you give
more context?

[mdempsky, 2014-09-09 22:52:12]

On 2014/09/09 20:29:18, Mark Seaborn wrote:
> I don't know the BPF DSL code very well.  What does this fix?  Could you give
> more context?

No other context really. It just makes the "using operator" statements
unnecessary in user code like nonsfi_sandbox.cc and
syscall_parameters_restrictions.cc.

[Mark Seaborn, 2014-09-10 02:11:49]

On 2014/09/09 22:52:12, mdempsky wrote:
> On 2014/09/09 20:29:18, Mark Seaborn wrote:
> > I don't know the BPF DSL code very well.  What does this fix?  Could you
give
> > more context?
> 
> No other context really. It just makes the "using operator" statements
> unnecessary in user code like nonsfi_sandbox.cc and
> syscall_parameters_restrictions.cc.

OK.  LGTM, though I don't really understand what the C++ lookup rules are doing
here.

[mdempsky, 2014-09-10 02:51:30]

On 2014/09/10 02:11:49, Mark Seaborn wrote:
> OK.  LGTM, though I don't really understand what the C++ lookup rules are
doing
> here.

Just in case you want a more detailed explanation...

Basically we have:

    template <typename T> class scoped_refptr { ... };

    namespace bpf_dsl {

    namespace interal {
    class BoolExprImpl { ... };
    }

    typedef scoped_refptr<internal::BoolExprImpl> BoolExpr;

    BoolExpr operator&&(BoolExpr, BoolExpr);

    } // namespace bpf_dsl

and suppose there's user code like:

    namespace nacl {
    using bpf_dsl::BoolExpr;

    void MyPolicy() {
      BoolExpr a = ..., b = ...;
      a && b;
    }
    }

If bpf_dsl::BoolExpr was a normal class, the "&&" operator would be resolved via
argument-dependent lookup by searching the "bpf_dsl" namespace, and the code
would compile fine.

However, ADL ignores typedefs when determining the set of associated namespaces,
and here bpf_dsl::BoolExpr is actually just a typedef for
scoped_refptr<bpf_dsl::internal::BoolExprImpl>.  Thus the only associated
namespace is bpf_dsl::internal.

That means when ADL tries to resolve "operator&&", it will check "::operator&&"
and "::bpf_dsl::internal::operator&&", but it won't check
"::bpf_dsl::operator&&", which is the function we actually declared/defined.

Adding the "using bpf_dsl::operator&&" declaration into namespace
bpf_dsl::internal then fixes this by making ::bpf_dsl::internal::operator&& into
an alias for ::bpf_dsl::operator&&.

[jln (very slow on Chromium), 2014-09-10 05:30:20]

lgtm

[mdempsky, 2014-09-10 05:33:41]

The CQ bit was checked by mdempsky@chromium.org

[commit-bot: I haz the power, 2014-09-10 05:34:57]

CQ is trying da patch. Follow status at
 https://chromium-status.appspot.com/cq/mdempsky@chromium.org/543333002/1

[commit-bot: I haz the power, 2014-09-10 06:48:49]

Committed patchset #1 (id:1) as 89312d0175abf5120b75102b3b724d86fd645273

[commit-bot: I haz the power, 2014-09-10 06:51:46]

Patchset 1 (id:??) landed as
https://crrev.com/b3038b5483c9601e64f05b3eb23e1a1f5b11e5ce
Cr-Commit-Position: refs/heads/master@{#294127}