[C++ Patch] PR 58126

[Paolo Carlini <paolo dot carlini at oracle dot com>]

Hi,

it seems to me that in order to fix this accepts-invalid (for the 
testcase we don't reject the declaration 'B b;') we have simply to 
propagate from bases to derived the CLASSTYPE_READONLY_FIELDS_NEED_INIT 
and CLASSTYPE_REF_FIELDS_NEED_INIT flags. Thus, the few class.c lines in 
the patch below. Then I think we can also make the error messages in 
diagnose_uninitialized_cst_or_ref_member_1 more clear and explicitly 
mention the base to which the problematic member belongs.

Tested x86_64-linux.

Thanks,
Paolo.

PS: I struggled a bit with "%qT" vs "%q#T": I would slightly prefer 
simply talking about, eg, "base class 'A'", we do that in other cases, 
but here we are already using # which automatically picks class vs 
struct. Thus I think we should use # for the base too... or remove all 
the #s and just talk about, eg, "class 'B'" and "base class 'A'" for 
these structs too?

//////////////

Attachment: CL_58126
Description: Text document

Attachment: patch_58126
Description: Text document