Persistent bugs in egcs 1.1/supplied libc++

[Claus Fischer <claus dot fischer at microworld dot com>]

Hi maintainers,
   Ulrich,

I checked egcs 1.1b for the four bugs I've reported.
Two of them are apparently persistent; most likely libc++ related.

When I reported them, I had apparently gotten too much heat or so; of the test
programs in my mails in the mailing list, three are wrong.  (Although I
had spent hours to reduce them to minimal form first.)

This time, if you find anything wrong with my programs, please shout back.

(1) -fno-default-inline
    With 1.0.3a I got duplicate symbols in my assembly file; with 1.1
    I get lots of error messages. However, they make much more sense
    and are perhaps related to 'bugs' in libc++, not in gcc.
    (should libc++ compile with -fno-default-inline?)

    ********************* p.cc **********************
       #include <iostream>

       class X
       {
       public:
       	   double x;
       };
       
       ostream& operator << (ostream & s,X  const & x)
       {
       	   s << x.x;
       	   return s;
       }
       
       int main() {};
    ************** end p.cc ************************

    compiled with /usr/local/egcs/bin/c++ -fno-default-inline -o p p.cc

(2) field witdth of strings
    This time clearly a libc++ bug.
    I remember reproducing it with char*'s instead of strings also;
    this behaviour is apparently gone.

    ******************* x.cc **************************
       #include <iostream>
       #include <iomanip>
       #include <string>
       
       int main() {
           string x = "hi there";
       	   cout << "!" << setw(10) << x << "!" << endl;
           cout << "!" << setw(10) << "hi there" << "!" << endl;
       }
    ***************** end x.cc ************************

    run with /usr/local/egcs/bin/c++ -o x x.cc; x
    and get
       !hi there         !
       !  hi there!
    with some blanks too much.
    Likely reason: internal length of string determined as 1 (like char).

(3) I wasn't able to reproduce the third of my outstanding bugs from July,
    regarding loop unrolling and -O1.
    (Again, the program included in my mail was not working, and I don't
    have the reduced form any more.)
    However, the bug seems gone cause I could compile my full program
    with -funroll-loops and -O3 now.

(4) The internal compiler error 9 that I've reported is fixed.



Non-functional test programs in three out of four bug reports is a pretty bad
statistics, and I can't really think that I've broken all three of them, on
different days.

Could there be some problem in the mailing list (archive), or was it my
mailer?  Perhaps the iso-accents-mode broke my cut-and-paste into Emacs.

System information: everything done on Linux,
   Linux boltzmann 2.0.33 #14 Sat Mar 7 17:14:37 MET 1998 i686 unknown
with libc 5.4.44 and this time egcs-1.1b.


Best regards

Claus