Re: GNU APL now requires bash shell and in-tree builds

[Dr . Jürgen Sauermann]

    Hi Alexander,
      
      thanks for offering help on this. I believe GNU APL cannot do much
      more than
      we already did (assuming the latest GNU APL now works again on
      BSD).
    
    I just thought that the BSD guys might want to look into this a
    little further because
    the code that the compiler now complains about is stone-old and
    worked
    just fine for more than 40 years. I suspect that the root cause of
    all this hides
    somewhere in the #include files and will, at some point in
    time, also affect
    other projects.
    
    i used gcc 8.4 in the past and I am almost sure that it works
    better. However,
    C compilers have gotten more and more picky lately and I wonder how
    long
    we will be able to keep up with the pace at which the compiler guys
    invent
    new problems.
    
    Best Regards,
    Jürgen

On 4/13/21 12:43 PM, Alexander Shendi wrote:

> Hi Jürgen,
>
> what I can try to do, is to compile GNU APL with g++ 8.4 from ports and not with clang++ 10 (system default).
>
> Best Regards,
>
> Alexander
>
> Am 13. April 2021 12:16:09 MESZ schrieb "Dr. Jürgen Sauermann" <mail@xn--jrgen-sauermann-zvb.de>:
> > Hi Brian,
> >
> > understood. However, sys/socket.h is directly #included in main.cc
> > (line 38).
> >
> > The question is then why bind() and, for example, accept(), which are
> > both declared in
> > sys/socket.h and are both used in main.cc in behave so differently as
> > to the need for ::
> > and only in BSD?
> >
> > And why would an extern "C" wrapper around it change the behaviour in
> > main.cc but is
> > not needed in all other *.cc files that happily use bind() without :: ?
> >
> > Looking at the error message posted by Alexander:
> >
> > main.cc:301:8: error: value of type '<const int &, const sockaddr *,
> > unsigned long>' is not contextually convertible to 'bool' if
> > (bind(listen_socket, (const sockaddr *)&local, sizeof(local)))
> > ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ then
> > apparently the compiler assumes that bind() does not return int (as
> > declared in sys/socket.h) but something else (maybe an enum?). Or maybe
> > it is a compiler fault? Best Regards,
> > Jürgen
> >
> >
> > On 4/13/21 2:54 AM, Callahan, Brian Robert wrote:
> >
> > Hi Jürgen --
> >
> >
> > Per POSIX, bind requires the inclusion of sys/socket.h, so that should
> > be in the includes list for main.cc if there is a call to bind in
> > main.cc. I'm not sure we can or should assume an implicit include from
> > elsewhere.
> >
> >
> > https://pubs.opengroup.org/onlinepubs/9699919799/functions/bind.html
> >
> >
> > As for needing the scope-resolution on bind, that is something that
> > someone (i.e., me or Alexander) should look further into.
> >
> >
> > ~Brian
> >
> >
> > _____________________________________________
> > Brian Robert Callahan, Ph.D.
> >
> > Lecturer, ITWS@RPI
> >
> > Office: Lally 304
> >
> > _____________________________________________
> >
> > From: Bug-apl [bug-apl-bounces+callab5=rpi.edu@gnu.org] on behalf of
> > Dr. Jürgen Sauermann [mail@xn--jrgen-sauermann-zvb.de]
> > Sent: Monday, April 12, 2021 5:47 AM
> > To: Alexander Shendi; bug-apl@gnu.org
> > Subject: Re: GNU APL now requires bash shell and in-tree builds
> >
> > Hi Alexander,
> >
> > thanks, fixed in SVN 1470.
> >
> > The bind() error is somewhat strange because:
> >
> > - in the same file (main.cc), accept() (a companion of bind()) works
> > without :: and
> > - in other files bind() also works without ::
> >
> > I suppose some other #include file declares bind() differently (which
> > would be an error in the platform).
> >
> > Best Regards,
> > Jürgen
> >
> >
> > On 4/11/21 10:53 PM, Alexander Shendi wrote:
> >
> > Dear list, I tried out SVN 1468, but still had to modify the following
> > files * . /configure (replace "source" with ".") * . /src/main.cc
> > (replace "bind" with "::bind" on line 300 Thanks in advance for your
> > help. /Alexander Am 11. April 2021 15:26:44 MESZ schrieb "Callahan,
> > Brian Robert" <callab5@rpi.edu>:
> >
> > Hello -- In fixing Alexander's issue, I found these lines in the GNU
> > APL Makefile I have never seen before: buildtag.hh: Makefile
> > buildtag.sh source ${top_srcdir}/buildtag.sh touch $@ The source
> > command is a bash built-in and won't work on other shells, like OpenBSD
> > ksh. Additionally, even with bash, this command does not work if doing
> > an out-of-tree build, which is the way I have always built GNU APL and
> > which autotools are supposed to make easy to do. Thanks. ~Brian
> > ________________________________ Brian Robert Callahan, Ph.D. Lecturer,
> > ITWS@RPI Office: Lally 304
> >
> > -- Ceci n'est pas un courriel.
> >       
> --
> Ceci n'est pas un courriel.
>