C++11 Compiling error on Windows

[Rodolfo Ochoa]

since new version of MS C++ Compiler is closer to C++11 standard, Windows is failing to compile due a bug in
 src/unit_tests/test_hashmaps.cpp
where a hash is created with zstring as key

[Paul J. Lucas (paul-lucas)]

It would help if you included the error message.

[Rodolfo Ochoa (rodolfo-ochoa)]

The error code doesn't say much:

Error 2 error C2338: The C++ Standard doesn't provide a hash for this type. (..\..\src\unit_tests\test_hashmaps.cpp) C:\Program Files (x86)\Microsoft Visual Studio 11.0\VC\include\xstddef 238 1 zorba_simplestore

but the error is on src\unit_tests\test_hashmaps.cpp:97:
std::unordered_map<zstring, int> map4(1024);

instead of zstring should be a primitive type.
std:string is allowed, but I think this will not fulfill the purpose of the test.

[Chris Hillery (ceejatec)]

This is preventing the trunk from compiling on Windows (at least with the latest compiler), so bumping the priority. Paul, please try to propose some solution ASAP.

[Paul J. Lucas (paul-lucas)]

Markos wrote the test_hashmaps.cpp file, not me. For *my* unit test file test_unordered_map.cpp, I correctly do *not* compile it if the C++ implementation provides its own unordered map.

This is Markos' bug, not mine.

[Paul J. Lucas (paul-lucas)]

Despite that, try adding:

#include "util/hash/hash.h"
#include "zorbatypes/zstring.h"

to the list of include files in test_hashmaps.cpp. I have no way to test this myself.

[Paul J. Lucas (paul-lucas)]

The other fix is simply not to use zstring and use std::string instead. He's presumably testing the hashmaps, not strings, so any string will do.

[Chris Hillery (ceejatec)]

This problem was introduced by change 11091 from Markos "Hashmap optimizations". (I still think Paul is probably the best person to look at it, but if Markos has a chance since Paul is out for the next week that would be good.)

I have pushed a branch which works around this problem by disabling test_hashmap:

  https://code.launchpad.net/~zorba-coders/zorba/windows-hashmap-workaround/+merge/131716

[Chris Hillery (ceejatec)]

Unfortunately Paul's suggestion in comment #5 doesn't fix it. The error changes to

test_hashmaps.cpp
C:\zorba\trunk\src\zorbaserialization/class_serializer.h(94) : error C2146: syntax error : missing ';' before identifier 't0'
C:\zorba\trunk\src\zorbaserialization/class_serializer.h(94) : error C4430: missing type specifier - int assumed. Note: C++ does not support default-int
C:\zorba\trunk\src\zorbaserialization/class_serializer.h(95) : error C2146: syntax error : missing ';' before identifier 't1'
C:\zorba\trunk\src\zorbaserialization/class_serializer.h(95) : error C4430: missing type specifier - int assumed. Note: C++ does not support default-int
C:\zorba\trunk\src\zorbaserialization/class_serializer.h(113) : error C2146: syntax error : missing ';' before identifier 'get_registration_time'
C:\zorba\trunk\src\zorbaserialization/class_serializer.h(113) : error C4430: missing type specifier - int assumed. Note: C++ does not support default-int
C:\zorba\trunk\src\zorbaserialization/class_serializer.h(113) : warning C4183: 'get_registration_time': missing return type; assumed to be a member function returning 'int'
C:\zorba\trunk\src\zorbaserialization/class_serializer.h(113) : error C2065: 't1' : undeclared identifier
C:\zorba\trunk\src\zorbaserialization/class_serializer.h(113) : error C2065: 't0' : undeclared identifier
C:\zorba\trunk\src\unit_tests\test_hashmaps.cpp(32) : warning C4244: 'return' : conversion from 'uint64_t' to 'uint32_t', possible loss of data
C:\zorba\trunk\src\unit_tests\test_hashmaps.cpp(101) : warning C4244: 'argument' : conversion from 'double' to 'float', possible loss of data
C:\zorba\trunk\src\unit_tests\test_hashmaps.cpp(102) : warning C4244: 'argument' : conversion from 'double' to 'float', possible loss of data
NMAKE : fatal error U1077: 'C:\PROGRA~2\MICROS~2.0\VC\bin\cl.exe' : return code '0x2'
Stop.

[Chris Hillery (ceejatec)]

Actually that appears to be the same error message as without Paul's suggestion.

[Chris Hillery (ceejatec)]

Even with my workaround, there are a number of compilation warnings that I do not believe were there prior to Markos' change. eg:

[ 21%] Building CXX object src/CMakeFiles/zorba_simplestore.dir/store/naive/store.cpp.obj
store.cpp
C:\zorba\trunk\src\zorbautils/hashmap.h(827) : warning C4244: 'return' : conversion from 'const double' to 'zorba::csize', possible loss of data
        C:\zorba\trunk\src\zorbautils/hashmap.h(826) : while compiling class template member function 'zorba::csize zorba::HashMap<T,V,C>::computeCapacity(zorba::csize) const'
        with
        [
            T=zorba::zstring,
            V=zorba::simplestore::XmlNode_t,
            C=zorba::HashMapZStringCmp
        ]
        C:\zorba\trunk\src\zorbautils/hashmap.h(395) : see reference to function template instantiation 'zorba::csize zorba::HashMap<T,V,C>::computeCapacity(zorba::csize) const' being compiled
        with
        [
            T=zorba::zstring,
            V=zorba::simplestore::XmlNode_t,
            C=zorba::HashMapZStringCmp
        ]
        c:\zorba\trunk\src\store\naive\store.h(181) : see reference to class template instantiation 'zorba::HashMap<T,V,C>' being compiled
        with
        [
            T=zorba::zstring,
            V=zorba::simplestore::XmlNode_t,
            C=zorba::HashMapZStringCmp
        ]

This leads me to believe there is a template-related error somewhere, which is why I think Paul is probably the best victim here.

[Markos Zaharioudakis (markos-za)]

The warnings from HashMap are easy to fix, and I will fix them in my next chakin. However, if zstring cannot be used as a key for std::unordered_map, that is Paul's problem, not mine. Or maybe zstring can be used as a key, but I am doing something wrong or incomplete in my code. However, if this is the case, I don't know what it is that I am doing wrong, so I would again ask Paul's help to fix it. So, I am reassigning this to Paul.

[Paul J. Lucas (paul-lucas)]

zstring can be used as a key, however, you didn't #include the file where the hash for zstring is in fact defined: util/hash/hash.h. Only #include'ing "unordered_map.h" is insufficient.

[Paul J. Lucas (paul-lucas)]

Also, as I originally said: you're presumably testing hashmaps, not zstrings, so there's no compelling reason for you to use zstring (as oppose to std::string) in your test.