Bug 113147 – Fix the build with GCC 4.8

• Status: RESOLVED
• Resolution: FIXED
• Component: Web Template Framework

[Jonathan Liu]

Compiling WebKit with GCC 4.8 generates lots of warnings due to COMPILE_ASSERT defining an unused typedef. Example:
generated/InspectorTypeBuilder.h: In member function 'WebCore::TypeBuilder::Network::CachedResource::Builder<STATE>::operator WTF::RefPtr<WebCore::TypeBuilder::Network::CachedResource>&()':
..\WTF/wtf/Assertions.h:328:47: warning: typedef 'dummyresult_is_not_ready' locally defined but not used [-Wunused-local-typedefs]
 #define COMPILE_ASSERT(exp, name) typedef int dummy##name [(exp) ? 1 : -1]
                                               ^
generated/InspectorTypeBuilder.h:2932:13: note: in expansion of macro 'COMPILE_ASSERT'
             COMPILE_ASSERT(STATE == AllFieldsSet, result_is_not_ready);

[Allan Sandfeld Jensen]

Compile_assert should probably be implemented using static_assert when available.

[Kai Koehne]

We could also just diable the warning:

#pragma GCC diagnostic ignored "-Wunused-local-typedefs"

(Note that you can't disable it 'locally' with pragma push, pragma pop though cause it's part of a #define).

[Allan Sandfeld Jensen]

(In reply to comment #1)
> Compile_assert should probably be implemented using static_assert when available.

Uploaded that to bug 113308

[Alexey Proskuryakov]

Does this bug still track anything useful, given that bug 113308 will fix the symptom?

[Allan Sandfeld Jensen]

(In reply to comment #4)
> Does this bug still track anything useful, given that bug 113308 will fix the symptom?

C++11 support may not be enabled in GCC. In fact it is default off.

[Zan Dobersek]

(In reply to comment #5)
> (In reply to comment #4)
> > Does this bug still track anything useful, given that bug 113308 will fix the symptom?
> 
> C++11 support may not be enabled in GCC. In fact it is default off.

The GTK port only enforces the C++11 standard when building the WebKit2 framework. This means that when building WTF, WebCore etc. static_assert is still not available and thus the same warnings are still being spat out.

Is this true for the Qt port as well?

[Andras Becsi]

(In reply to comment #6)
> (In reply to comment #5)
> > (In reply to comment #4)
> > > Does this bug still track anything useful, given that bug 113308 will fix the symptom?
> > 
> > C++11 support may not be enabled in GCC. In fact it is default off.
> 
> The GTK port only enforces the C++11 standard when building the WebKit2 framework. This means that when building WTF, WebCore etc. static_assert is still not available and thus the same warnings are still being spat out.
> 
> Is this true for the Qt port as well?

As far as I know Qt does not enforce C++11 on any framework yet.

[Allan Sandfeld Jensen]

(In reply to comment #6)
> (In reply to comment #5)
> > (In reply to comment #4)
> > > Does this bug still track anything useful, given that bug 113308 will fix the symptom?
> > 
> > C++11 support may not be enabled in GCC. In fact it is default off.
> 
> The GTK port only enforces the C++11 standard when building the WebKit2 framework. This means that when building WTF, WebCore etc. static_assert is still not available and thus the same warnings are still being spat out.
> 
> Is this true for the Qt port as well?

Let me put it this way: WebCore doesn't build with C++11 enabled, see bug 111579.

[Zan Dobersek]

Right, understood.

The warnings that the dummy COMPILE_ASSERT implementation is throwing out can be suppressed using the __attribute__((unused)) directive. It seems to work with GCC 4.7 and 4.8 and with Clang 3.2, though there are other compilers that people use that don't necessary support the directive.
https://svn.boost.org/trac/boost/ticket/7242

[Jonathan Liu]

Created attachment 195890 [details]
Patch

[WebKit Review Bot]

Attachment 195890 [details] did not pass style-queue:

Failed to run "['Tools/Scripts/check-webkit-style', '--diff-files', u'Source/WTF/ChangeLog', u'Source/WTF/wtf/Assertions.h']" exit_code: 1
Source/WTF/wtf/Assertions.h:354:  Extra space before [  [whitespace/braces] [5]
Total errors found: 1 in 2 files


If any of these errors are false positives, please file a bug against check-webkit-style.

[Raphael Kubo da Costa (:rakuco)]

There are also warnings generated by

Source/WTF/wtf/dtoa/utils.h:294:22: error: typedef ‘VerifySizesAreEqual’ locally defined but not used [-Werror=unused-local-typedefs]
         typedef char VerifySizesAreEqual[sizeof(Dest) == sizeof(Source) ? 1 : -1];

Source/WTF/wtf/HashTable.h:1302:84: warning: typedef ‘HashTableType’ locally defined but not used [-Wunused-local-typedefs]
         typedef HashTable<Key, Value, Extractor, HashFunctions, Traits, KeyTraits> HashTableType;
                                                                                    ^

Source/WTF/wtf/HashTable.h:1303:97: warning: typedef ‘const_iterator’ locally defined but not used [-Wunused-local-typedefs]
         typedef HashTableConstIterator<Key, Value, Extractor, HashFunctions, Traits, KeyTraits> const_iterator;
                                                                                                 ^
Doesn't it make more sense to add a macro to Compiler.h that's defined when we have either gcc or clang and use it in all those declarations?

[Anders Carlsson]

This doesn't seem like a helpful warning, let's just turn it off.

[Raphael Kubo da Costa (:rakuco)]

(In reply to comment #13)
> This doesn't seem like a helpful warning, let's just turn it off.

Hmm, do you mean leaving it up to each build system to detect the compiler and its version and pass -Wno-unused-local-typedefs accordingly?

[Allan Sandfeld Jensen]

(In reply to comment #14)
> (In reply to comment #13)
> > This doesn't seem like a helpful warning, let's just turn it off.
> 
> Hmm, do you mean leaving it up to each build system to detect the compiler and its version and pass -Wno-unused-local-typedefs accordingly?

I think it would be better to disable it in source code than to try to rely on specific compiler flags.

[Raphael Kubo da Costa (:rakuco)]

(In reply to comment #15)
> (In reply to comment #14)
> > (In reply to comment #13)
> > > This doesn't seem like a helpful warning, let's just turn it off.
> > 
> > Hmm, do you mean leaving it up to each build system to detect the compiler and its version and pass -Wno-unused-local-typedefs accordingly?
> 
> I think it would be better to disable it in source code than to try to rely on specific compiler flags.

I agree, and I think this was the direction we were moving towards, but Anders' comment sounded like he had a different opinion.

[Andras Becsi]

Created attachment 196483 [details]
Patch

[Allan Sandfeld Jensen]

Comment on attachment 196483 [details]
Patch

LGTM

[Andras Becsi]

Comment on attachment 196483 [details]
Patch

Clearing flags on attachment: 196483

Committed r147640: <http://trac.webkit.org/changeset/147640>

[Andras Becsi]

All reviewed patches have been landed.  Closing bug.