RE: regression in qmtest_libstdcpp

 
>>Reviewing the gcc-4.0.1 source, I see the upstream tests have 
>changed, ...
>>
>>-  verify_demangle("f", "error code = -2: invalid mangled name");
>>+  verify_demangle("f", "float");
>>
>>Making the above change in the qmtest source tree and re-running 
>>build_binary_testsuite, they now pass in that environment (they were 
>>failing in the same manner as the binary test suite)
>>
>>Testing against our 10.2 release with libstdc++.so.6 built 
>>from gcc-3.4.3, the tests pass.

Stew confirmed on IRC that the updated test fails against
an old library, which wasn't a surprise, but good to know.

>The test changelog entry is from Feb 15, and says "Adjust for 
>2005-02-13 demangler change".  I don't actually see a changelog
>entry in gcc that would indicate such a change, nonetheless,
>the date on this is after the 3.4.3 release, but before the 
>3.4.4, 4.0, or 4.0.1 releases.

Okay, I found the entry, it was in libiberty where I didn't
think to look first.  This is from the 3.4 branch:

2005-05-19  Release Manager

	* GCC 3.4.4 released.

2005-02-13  Jason Merrill  <[email protected]>

	* cp-demangle.c (__cxa_demangle): Change resolution of ambiguous
	arguments.

2004-11-04  Release Manager

	* GCC 3.4.3 released


Here are the various references:

// IA 64 C++ ABI - 5.1 External Names (a.k.a. Mangling)

#include <testsuite_hooks.h>

// Examples given in the IA64 C++ ABI 
// http://www.codesourcery.com/cxx-abi/abi-examples.html#mangling

Which says:

f	C function or variable "f" or a global namespace variable "f" 

Working off the spec wording which says:

5.1.2 General Structure 

Entities with C linkage and global namespace variables are not mangled.