[RFC] Dealing with ODR violations in GCC

[Jan Hubicka]

Hi,
the warning about types of fields seems misplaced:

> ../../third_party/WebKit/public/platform/WebCryptoAlgorithmParams.h:295:0: warning: type ???struct WebCryptoEcKeyImportParams??? violates one definition rule [-Wodr]
>  class WebCryptoEcKeyImportParams : public WebCryptoAlgorithmParams {
>  ^
> ../../third_party/WebKit/public/platform/WebCryptoAlgorithmParams.h:295:0: note: a different type is defined in another translation unit
>  class WebCryptoEcKeyImportParams : public WebCryptoAlgorithmParams {
>  ^
> ../../third_party/WebKit/public/platform/WebCryptoAlgorithmParams.h:307:0: note: the first difference of corresponding definitions is field ???m_namedCurve???
>      const WebCryptoNamedCurve m_namedCurve;
>  ^
> ../../third_party/WebKit/public/platform/WebCryptoAlgorithmParams.h:295:0: note: a field of same name but different type is defined in another translation unit
>  class WebCryptoEcKeyImportParams : public WebCryptoAlgorithmParams {
>  ^
> ../../third_party/WebKit/public/platform/WebCryptoAlgorithm.h:78:0: note: type ???const WebCryptoNamedCurve??? should match type ???const WebCryptoNamedCurve???
>  enum WebCryptoNamedCurve {
>  ^
> ../../third_party/WebKit/public/platform/WebCryptoAlgorithm.h:79:0: note: the incompatible type is defined here
>      WebCryptoNamedCurveP256,
>  ^

Not sure how it can happen, but will take a look.  The third note should not point to the type itself but to the field.

> ../../third_party/icu/source/i18n/unicode/fieldpos.h:106:0: warning: virtual table of type ???struct FieldPosition??? violates one definition rule  
>  class U_I18N_API FieldPosition : public UObject {
>  ^
> ../../third_party/icu/source/i18n/unicode/fieldpos.h:106:0: note: the conflicting type defined in another translation unit
>  class U_I18N_API FieldPosition : public UObject {
>  ^
> ../../third_party/icu/source/i18n/unicode/fieldpos.h:106:0: note: virtual method ???_ZTIN6icu_5213FieldPositionE???
>  class U_I18N_API FieldPosition : public UObject {
>  ^
> ../../third_party/icu/source/i18n/format.cpp:125:0: note: ought to match virtual method ???__comp_dtor ??? but does not
>  Format::format(const Formattable& /* unused obj */,
>  ^

This should warn about different number of entries in the table rather than insisting that typeinfo should match destructor.
Jason, I wonder if there is more informative way to print destructor during LTO than as __comp_dtor? Laos for named types,
perhaps printing just type name instead of the wrong "struct typename" (because we make no difference between struct and
class at LTO) would be cleaner.

Honza