new testsuite failure

[Gabriel Dos Reis]

Phil Edwards <pedwards@disaster.jaj.com> writes:

| The 27_io/istream_seeks now fails to build, I think as a result of the recent
| "should we deprecate <some extension> or not?" discussion on the main lists.
| The following message is printed many times, and then compilation just
| /stops/ with no further messages or indication as to what actually failed.
| 
| /home/pme/src/unified/libstdc++-v3/testsuite/27_io/istream_seeks.cc: In
|    function `void test04()':
| /home/pme/src/unified/libstdc++-v3/testsuite/27_io/istream_seeks.cc:115: choosing
|    `std::fpos<_State> std::operator+(const std::fpos<_State>&, long int) [with
|    _State = mbstate_t]' over `operator+'
| /home/pme/src/unified/libstdc++-v3/testsuite/27_io/istream_seeks.cc:115:
|    because worst conversion for the former is better than worst conversion for
|    the latter
| 
| That text reads like an 'informative' message rather than an error.

which I consider is a Good Thing :-)

| Hmmm... is there an -fdiagnostics switch to trim the dirname from the
| source file when it is displayed?

Not yet...

| ...  Perhaps something akin to the -p switch
| to patch(1), optionally specifying the number of directories to keep?
| The line that ends with "choosing" still wraps on my screen when this runs.

Yes, that is ugly. Actually the Right Thing to do would be to use a
single call to cp_error() -- or error() later -- instead of multiple
calls to cp_error(), cp_error_at() to print a single diagnostic.  The
main reason there are multiple calls is that each cp_error() is
specifying different instruction as to which line number to use in
diagnostic reports.  That is fragile.  Ideally, multiple format
specifiers affecting line numbers should be permitted, and the
diagnostic library should be able to pick the right one.  There were
some discussions to that effect between me and Nathan Sidwell, but
that is not a trivial issue.

-- Gaby
CodeSourcery, LLC                       http://www.codesourcery.com