Boost C++ Libraries: Ticket #11599: VS2015 warns that in new_tree, not all control paths return a value

anonymous — Wed, 28 Oct 2015 21:27:46 GMT

This warning occurs in VS2010 as well. This becomes an even bigger problem when you treat warnings as errors. Can you put in something to fix this please?

anonymous — Tue, 08 Dec 2015 16:28:29 GMT

Same here!

thanks

alister_fong@… — Wed, 23 Dec 2015 02:20:36 GMT

This appears for VS2010 using Boost 1.60 as well. Thank you.

stephan.menzel@… — Mon, 29 Feb 2016 14:16:41 GMT

Confirmed. Also the case for MSVC14 and 1.60.

status changed; resolution set

Sebastian Redl — Tue, 01 Mar 2016 15:31:26 GMT

status new → closed
resolution → fixed

In my opinion, VS's warning is broken and I would have loved to be able to use Clang's uncovered-switch warning. But I've bit the sour apple and did this because I couldn't find any nicer way of silencing VS without outright suppressing the warning (and thus adding compiler-specific code).

anonymous — Wed, 11 May 2016 08:42:20 GMT

Replying to cornedbee:

In my opinion, VS's warning is broken and I would have loved to be able to use Clang's uncovered-switch warning. But I've bit the sour apple and did this because I couldn't find any nicer way of silencing VS without outright suppressing the warning (and thus adding compiler-specific code).

I think this is an error because assert does nothing if NDEBUG is defined.

JM Volle — Wed, 11 May 2016 11:47:18 GMT

In case it helps, what I did (warning seen with VS2015) to suppress the warning is simply add:

diff --git a/LIB/CPP_3rdParty/boost/1_60/boost/property_tree/detail/json_parser/standard_callbacks.hpp b/LIB/CPP_3rdParty/boost/1_60/boost/property_tree/detail/json_parser/standard_callbacks.hpp
index 56c378e..bfa5374 100644
--- a/LIB/CPP_3rdParty/boost/1_60/boost/property_tree/detail/json_parser/standard_callbacks.hpp
+++ b/LIB/CPP_3rdParty/boost/1_60/boost/property_tree/detail/json_parser/standard_callbacks.hpp
@@ -129,6 +129,8 @@ namespace boost { namespace property_tree {
                 return new_tree();
             }
             assert(false);
+            throw(std::logic_error("reached passed assert!"));
+
         }
         string& new_value() {
             if (stack.empty()) return new_tree().data();
--

Wed, 11 May 2016 18:33:56 GMT

Adding a throw with a custom string runs mildly contrary to the C++ tradition of "don't pay for what you don't use". In this case, everyone knows that the code will never be called - but not the compiler, which will do its duty to use the necessary code space, with its accompanying penalties for load time and cache hit rate. Any thought of my original suggestion of just making the one of the cases be the default?

stephan.menzel@… — Wed, 11 May 2016 18:38:33 GMT

Well, wouldn't it sort of achieve the desired effect to make object: the default?