Re: [boost] [serialization] test_variant_xml_archive - Borland 5.82

From: Robert Ramey (ramey_at_[hidden])
Date: 2006-12-04 12:18:51

Vladimir Prus wrote:
> Nicola Musatti wrote:
>
>> Robert Ramey wrote:
>>  > Vladimir Prus wrote:
>>  >> Robert,
>>  >> for all appearances, this test failure:
>>  >>
>>  >>        http://tinyurl.com/y5gv7s
>>  >>
>>  >> is the problem with the test itself, not with the library. The
>>  >> error message is:
>>  >>
>>  >>   Error E2304 ..\libs\serialization\test\test_variant.cpp 171:
>>  >>   Constant variable '_Stz' must be initialized
>>  >>
>>  >> The problem is that I can't find that many lines in the file in
>>  >> question. Do you have an idea what's this variable is and how to
>>  fix >> it?
>>  >
>>> I've seen this and am not sure how to respond.  Given that it
>>> points to a line#
>>> beyond the end-of-file its looks to me that its an issue with
>>> intialization of some static variable with some library - probably
>>> with the
>>> standard library implementation for this environment.  Worse yet,
>>> it seems to come and go.  In any case, the best person to
>>> investigate this woudl be Nicholas Musetti as he has made
>>> the changes required to support this compiler.
>>
>> I believe this refers to std::fpos<>::_Stz from <iosfwd>. This
>> appears to refer to a specialization on a const type, which in turn
>> makes me think of Borland's bug of not dropping the top level const
>> qualification in argument deduction.
>>
>> What I do not know is what is triggering this and whether in this
>> case the usual workaround is applicable, which is to specify
>> explicitly the relevant function's template arguments.
>
> Would it be possible for you to prepare this "usual workaround" and
> see if
> it helps?
>

This really should be part of the variant library, I don't know what the
usual workaround is, I don't have a copy of this compiler, and if the
above speculartion is correct, its likely just a test artifact anyway.
Nicolas has spent a lot of time finding the fixes required for borland
5.82 and I think that's enough.

.How about just marking it as failing and move on.

Robert Ramey