Re: [Log4cpp-devel] Porting log4cpp to QNX Neutrino 6.1

[Bastiaan B. <Bas...@li...>]

On Wed, 2002-09-04 at 10:16, Harald Wellmann wrote:
>  Bastiaan Bakker  schrieb am 03.09.02 18:34:55:
> > On Sat, 2002-08-24 at 12:53, Harald Wellmann wrote:
> > >
> > > the short story is: log4cpp is running on QNX 6.1 without significant
> > > changes, and it passes all tests.
> > >
> >
> > That's good to hear. Which version are you talking about? 0.2.7 or
> > 0.3.x?
> >
> 
> I've been working on the 0.3.2rc4 tarball.
> 
> 
> > > Now here's a summary of the changes I had to make:
> > >
> > > - With the -pedantic compiler option, gcc gets so pedantic it refuses to
> > > compile anything at all, so I turned it off.
> > >
> > I believe I already turned off -pedantic for g++ 2.96 to get rid of this
> > endless stream of warnings about code in the iostreams lib. Maybe I
> > should turn it off completely by default and only enable it for testing.
> >
> 
> Yes, I've found that in m4/PETI_PEDANTIC_GCC.m4. It's turned off for
> 2.96*, but 2.95* is covered by the default case. We could simply add
> another case for 2.95.
> 
Yes. I'll switch main log4cpp development to gcc 3.2 soon anyway, so
being less pedantic for older gcc version would be wise.

> 
> > > - Optimization -O2 produces internal compiler errors on some sources, so
> > > I removed that as well.
> > >
> 
> Is that m4 file also the "right" place to disable -O2 for QNX 6.1.0?
> 

Well, if we promote this macro to a generic 'tune gcc' thing, then it's
the right place. Can you check what the right platform identifier is for
QNX? Then we can add the appropiate check. Run config/config.guess to
obtain the identifier.
 

> 
> > > - gcc by default simply ignores the std:: namespace (at least in
> > > versions <= 2.95.3; I don't know about the most recent versions), but
> > > not in the QNX variant with the Dinkum libs. In most files, I had to
> > > insert "using namespace std;" or prefix a few function calls with std::.
> > >
> >
> > g++ < 3.0 is lax in enforcing proper namespace usage. However, 3.0 and
> > higher do complain, so any place where log4cpp is missing a necessary
> > 'std::' is a bug that should be fixed. Please report any occurrences you
> > encounter. BTW, I prefer to avoid 'using namespace ..' because it tends
> > to defeat the purpose of namespaces.
> >
> 
> Yes, I agree. And since "using std::map;" (i.e. single identifiers instead
> of the
> complete namespace) does not work on WinCE 3.0, I'd better use explicit
> prefixes for all occurrences.
> 
> 
> > > - A similar problem is the syntax of standard C header includes. On QNX,
> > > you have to write #include instead of #include , or
> > > you will get lots of symbol redefinition errors.
> > >
> >
> > Yes, we'll have to change these as well, instead of
> > is the 'proper' C++ way.
> >
> 
> Exactly, and <> brackets are reserved for system includes, so you we should
> never write #include but use "" instead. Someone has started
> cleaning that up between 0.2.* and 0.3.* but you still see the angular
> brackets
> in a few places.
>

This is a misconception. <> brackets are not reserved for system
includes. The only thing the ISO C++ standard says (in par. 16.2) about
the differences between "" and <> is that for "" the preprocessor will
search in 'an implementation dependent manner' and will fall back to
searching like a #include <> if this fails. Gcc has implemented this as
searching in the directory of the original source file for "" and
searching in the standard directories (/usr/include, etc.) and those
specified with '-I' for <>. This is also what Stroustrup specifies in
his reference manual (par. 16.3.2c).
Therefore the convention in log4cpp is: if a header file is part of its
interface, that is, its in include/log4cpp/ or subdirectory include it
with #include<>. If on the other hand it's internal (e.g. in src/)
#include it with #include""
 
 
> > > - Some of the global Makefile settings are different, of course. The
> > > compiler and linker is called QCC, and you don't link any external
> > > libraries. pthread_* stuff is available by default.
> >
> > The compiler and linker can be changed by setting environment variables
> > CC and LD prior to invoking ./configure. E.g. with the bash shell:
> >
> > CC=QCC LD=QCC ./configure --prefix=...
> >
> > The pthread detection M4 macro has been rather quickly hacked together
> > to get pthread support in. We should be able check we need -lpthread or
> > not for pthread support.
> > If you remove the '-lpthread' from 'LIBS="$LIBS -lpthread"' in
> > ./configure, does ./configure still recognize the pthread support?
> >
> 
> I'll try that and let you know the results.
> 
> 
> > >
> > > - libtool refused to produce libraries. I installed a newer libtool
> > > version and I got archived libraries, but I haven't found out how to
> > > produce shared libs yet.
> > >
> >
> > OK, so '--disable-shared' is recommended for now? Something to put in
> > the build documentation.
> >
> 
> It works without --disable-shared, libtool simply prints a warning that it
> won't generate shared libraries. Anyway, the problem is somewhere else,
> because shared libraries as such are no problem at all on QNX.
> 
> I don't know enough about libtool, but my guess it that it might work
> if I installed the newest autoconf/automake/libtool stuff and rerun
> autoconf and automake.

I use autoconf 2.53, automake 1.6.2 and 1.4.2. Should be new enough! 
Maybe the libtool FAQ or mailinglist have some clues...
 
> 
> I don't understand how the libtool generated by configure in the log4cpp
> root directory relates to the libtool installed on your platform, but simply
> using the installed one instead of the generated one (that what I did to
> make it work at all) is not the best way, I suppose.

If you don't run ./autogen.sh before running ./configure (and for a
log4cpp from a tar ball you don't need to), the libtool installed on
your platform won't be used at all. 

> 
> So it seems I'm going to learn more about autoconf and friends than I
> ever wanted to know...
>

Heheh, before you know it you're sucked into the tar pit of M4
bastardizations :-)

Regards,

Bastiaan

 
> Regards,
> Harald
> 
> -- 
> GMX - Die Kommunikationsplattform im Internet.
> http://www.gmx.net
> 
> 
> 
> -------------------------------------------------------
> This sf.net email is sponsored by: OSDN - Tired of that same old
> cell phone?  Get a new here for FREE!
> https://www.inphonic.com/r.asp?r=sourceforge1&refcode1=vs3390
> _______________________________________________
> Log4cpp-devel mailing list
> Log...@li...
> https://lists.sourceforge.net/lists/listinfo/log4cpp-devel