Address bind_handler ambiguous overload error when boost/bind/std_placeholders.hpp is included

[ednolan]

Previously, the implementation of bind_handler made the assumption that the trait boost::is_placeholder would be false for types corresponding to values in the std::placeholders namespace.

However, boost/bind commit c85b31e3d200dda2a73cf0027a82c6d8e29066f8, Support use of standard placeholders with boost::bind, added a new header, boost/bind/std_placeholders.hpp, which adds specializations to
boost::is_placeholder for std::placeholders types.

When this header is included before a use of boost::bind_handler, it results in compiler errors like the following, due to an internal helper function having overloads for
enable_if<boost::is_placeholder...> and
enable_if<std::is_placeholder...>, which were previously assumed to be mutually exclusive, but for which both conditions now are true:

../../../boost/beast/core/detail/bind_handler.hpp:126:18: error: call of overloaded 'extract(std::_Placeholder<1>, boost::beast::detail::tuple<int&&>)' is ambiguous
  126 |         h(extract(detail::get<S>(std::move(args)),
      |           ~~~~~~~^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  127 |             std::forward<ValsTuple>(vals))...);
      |             ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

This header is included in boost/bind/bind.hpp, and is transitively included by headers in many boost libraries including:

algorithm
asio
atomic
graph
msm
multi_index
property_tree
ptr_container
python
signals2
test
thread

Making it possible that an include from one of these libraries will randomly cause bind_handler to fail to compile.

This change addresses the issue by ensuring that boost/bind/std_placeholders.hpp is included in the bind_handler implementation file, and adjusting the SFINAE for the boost::is_placeholder overload so that it triggers only
if std::is_placeholder is false. This approach is necessary over simply removing the std::is_placeholder overload because std_placeholders.hpp might have had its functionality suppressed by feature guards. This change also explicitly adds a boost/bind/std_placeholders.hpp include to the bind_handler unit test file to prevent this issue from regressing.

[vinniefalco]

Thank you for your contribution!

[cppalliance-bot]

Timeline tracing charts: https://2638.beast.tracing.cppalliance.org/index.html

[codecov]

Codecov Report

Merging #2638 (98b681b) into develop (341ac75) will not change coverage.
The diff coverage is n/a.

Additional details and impacted files

@@           Coverage Diff            @@
##           develop    #2638   +/-   ##
========================================
  Coverage    92.93%   92.93%           
========================================
  Files          177      177           
  Lines        13651    13651           
========================================
  Hits         12687    12687           
  Misses         964      964           

Impacted Files	Coverage Δ
include/boost/beast/core/detail/bind_handler.hpp	100.00% <ø> (ø)

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 341ac75...98b681b. Read the comment docs.

[klemens-morgenstern]

Looks good; but can't we just include the std_placeholders.hpp and just add one function? Or do you want to keep the headers to a minumum @vinniefalco ?

[vinniefalco]

That's up to you. Don't break anything.

[ednolan]

can't we just include the std_placeholders.hpp and just add one function

If what you mean by this is, can't we have a single overload handling the std::placeholders and boost::placeholders cases: that was my original approach. But the problem is that std_placeholders.hpp isn't guaranteed to actually work for std::placeholders because it can be turned off by feature gating. This caused my original commit to fail CI for GCC 4. So you still need two overloads.