[cfe-dev] Patch to correct compilation errors related to class friend declarations

[John McCall]

On Nov 6, 2011, at 6:47 PM, Tom Honermann wrote:

> On 11/5/2011 3:05 AM, John McCall wrote:
>> On Nov 4, 2011, at 8:34 PM, Tom Honermann wrote:
>>> I checked out cfe/trunk today and attempted a build using an older
>>> gcc compiler (4.0.3) on Linux.  The build failed in a few places
>>> with the errors below.  The attached patch corrects these.  Each of
>>> the errors was related to class friend declarations.
>>> 
>>> llvm-trunk/tools/clang/include/clang/Basic/Diagnostic.h:590: error: 'clang::DiagnosticsEngine::<anonymous enum> clang::DiagnosticsEngine::MaxArguments' is private
>>> llvm-trunk/tools/clang/include/clang/Basic/PartialDiagnostic.h:36: error: within this context
>>> llvm-trunk/tools/clang/include/clang/Basic/Diagnostic.h:621: error: 'clang::DiagnosticsEngine::<anonymous enum> clang::DiagnosticsEngine::MaxFixItHints' is private
>>> llvm-trunk/tools/clang/include/clang/Basic/PartialDiagnostic.h:68: error: within this context
>>> 
>>> This error looks legit to me.  The code appears to be violating
>>> C++03 11.4.2 [class.friend] and 11.8.1 [class.access.nest].
>>> Granting friendship to a class does not confer friendship to nested
>>> classes of the befriended class.
>> 
>> It does, actually: [class.access]p2: A member of a class can also
>> access all the names to which the class has access.
>> [class.access.nest]p1: A nested class is a member and as such has the
>> same access rights as any other member. It's the other way that
>> doesn't hold;  having access to a class doesn't give you access to
>> its nested classes.
> 
> It looks like you are quoting from C++11 (or draft) and that the behavior has changed from C++03 (or, there was a CWG defect related to this?).

Yes, this was DR 45, with proposed resolution as stated in 2001.  It didn't make it into C++03 because <bureaucracy>.
  http://www.open-std.org/jtc1/sc22/wg21/docs/cwg_defects.html#45

John.