QGIS » QGIS Application

I noticed this problem when building Lib_Refactoring-branch branch.

BTW, is this possible to add branches selection to Ticket Properties?

I'd like to point one more resource with explanation of this problem:

When standards collide: the problem with dlsym

This issue is also a subject of C++ Standard Core Language Defects:

195. Converting between function and object pointers

Finally, I'd suggest to close this bug with appropriate comment or mark as enhancement for future, when C++ Standard and implementations (compilers/libraries) are fixed.

Fortunately, it's not a bug in QGIS itself.

Changed by anonymous?!?

Please, anyone, log-in if are attempting to change status of tickets.
Otherwise it's unknown who to talk to about a ticket, if needed.

I'd suggest we mark it as an enhancement, and set the milestone to blank. That way it hangs around but doesn't show up in the 'tickets left in milestone' totals.

Perfect. I agree.
I believe the state should be changed by assignee.

The assignee is the default person for Build/Install tickets, so Gary probably hasn't explicitly asked for this one :)

Actually, not even reinterpret_cast will work without warnings. One way to get around this (the behaviour is still undefined, strictly speaking, but at least the warnings go away) is to define a template function like this:

template <typename TO, typename FROM> TO nasty_cast(FROM f) {
union {
FROM f;
TO t;
} u;
u.f = f;
return u.t;
}

and then call it like this:

void* foo = something();
  my_func_ptr = nasty_cast<MyFuncPtrType>(foo);

Head builds without warnings now - we need to check if this report can be closed now. I reset the milestone to 0.9 since from now on we build with -Wall -Werror so if it exists in trunk it should be treated as a bug

• blocker - bugs that should block the release. Since we are going to release pretty much 'come what may' I would like no bugs
• allocated to this category without consultation with me and / or PSC
• critical - bugs that cause the application to crash or corrupt data
• major - application features that do not function at all
• minor - features that function but imerfectly e.g. labels placing incorrectly
• trivial - gui useability issues or small issues with the documentation, install notes etc.

Tim,

Current version of QGIS SVN trunk does not set any flags like -Wall, -Werroretc. I have to do it manually:

cmake ... -D CMAKE_CXX_FLAGS:STRING="-pedantic -Wall -Wno-long-long -fstrict-aliasing -Wstrict-aliasing=1" ..

-Werror flag removed, becuase the baby does not build with it, warnings occur.

@anonymous

Agreed, because nasty_cast does the same what std::memcpy can do.

Tim,

Continuing my previous comment, current SVN trunk does still thrown the same warning if compiled with my set of CXX flags:

trunk/qgis/src/core/qgsproviderregistry.cpp: In constructor 'QgsProviderRegistry::QgsProviderRegistry(QString)':
trunk/qgis/src/core/qgsproviderregistry.cpp:115: warning: ISO C++ forbids casting between pointer-to-function and pointer-to-object
trunk/qgis/src/core/qgsproviderregistry.cpp:119: warning: ISO C++ forbids casting between pointer-to-function and pointer-to-object
trunk/qgis/src/core/qgsproviderregistry.cpp:127: warning: ISO C++ forbids casting between pointer-to-function and pointer-to-object
trunk/qgis/src/core/qgsproviderregistry.cpp:128: warning: ISO C++ forbids casting between pointer-to-function and pointer-to-object
trunk/qgis/src/core/qgsproviderregistry.cpp:138: warning: ISO C++ forbids casting between pointer-to-function and pointer-to-object
trunk/qgis/src/core/qgsproviderregistry.cpp:146: warning: ISO C++ forbids casting between pointer-to-function and pointer-to-object
trunk/qgis/src/core/qgsproviderregistry.cpp: In member function 'QgsDataProvider* [[QgsProviderRegistry]]::getProvider(const QString&, const QString&)':
trunk/qgis/src/core/qgsproviderregistry.cpp:353: warning: ISO C++ forbids casting between pointer-to-function and pointer-to-object

Folks, as I see dlsym is used indirectly but through QLibrary::resolve. The http://doc.trolltech.com/4.0/qlibrary.html#resolve Qt documentation] says:

The symbol must be exported as a C function from the library.

Does it happen in QGIS? In fact, it would solve the problem I've reported here.

Union trick, memcpy and other hacks do not, as C++ FAQ says: It's illegal, period.

Jef,

I agree about where is the root of the problem and extern "C" does not solve all aspects of it. Clean and 100% valid solution for this problem does not exist. So, silencing warning does not eliminate it, potential risk of problems on various C++ implementations might be expected.

Milestone Version 1.0.0 deleted

Warming up an old steak, I've learned that the union-based cast used in cast_to_fptr is not necessary. GCC accepts reinterpret_cast for purposes like in dlsym calls as an extension:

#ifdef +GNUC+
+extension+
#endif
isprovider_t *hasType = reinterpret_cast<isprovider_t*>(myLib->resolve("type"));