Re: c++/6001: g++ problems with nested headers from > egcs-2.91.66 until 3.0.0

[Neil Booth <neil@daikokuya.demon.co.uk>]

The following reply was made to PR c++/6001; it has been noted by GNATS.

From: Neil Booth <neil@daikokuya.demon.co.uk>
To: Mario Deilmann <Mario.Deilmann@pallas.com>
Cc: gcc-gnats@gcc.gnu.org
Subject: Re: c++/6001: g++ problems with nested headers from > egcs-2.91.66 until 3.0.0
Date: Mon, 25 Mar 2002 18:26:25 +0000

 Mario Deilmann wrote:-
 
 > You never should say there are no bug in a specific software ;).
 
 I'm quite sure there are no bugs in this part of CPP.  People use it in
 millions of situations every day.
  
 > > You should investigate what the issue is.  Have you created preprocessed
 > > source and examined it?  Have you created preprocessed source with a
 > > compiler that works and examined it?  I bet there's a difference, and we
 > > need to know what is causing it.
 > 
 > I did some investigations. Before I send you the bug report I tested the
 > problem with various compilers on different platforms and what I observed
 > was that the error only occured with versions of gcc. Then I modified the
 > source only by changing the include sequence and after this all gcc
 > versions had no problem any more. I want to tell the gcc people about this
 > so I send the bug report. I have of course limited time to investigate
 > on a problem where I have a workaround for and the different include
 > sequences I generated with your help don't tell anything to me and I
 > definetly don't have the time to look in the cpp source code. Sorry !
 
 I'm not asking you to look into the cpp source code.  I'm asking you
 for *preprocessed source*.  You haven't sent it.  It's the output of -E.
  
 > You don't need to replicate my system. With the source code example I send
 > you you could replicate the error on a lot of systems.
 
 I can't, on g++ on Linux I get:
 
 $ g++ -c -Wall buffer.cpp 
 buffer.cpp: In method `char TSourceBuffer::GetLine()':
 buffer.cpp:137: warning: unused variable `int lineNumber'
 $ 
 
 in both directories error/ and no-error/ of your tarball.
 
 > I only tested 4 but
 > I'm sure there are a lot more. The problem occured only with versions of
 > gcc (and to be honest with the Intel Compiler on Linux too) but all other
 > C++ Compiler I used didn't find a problem or even a bug (I tested CC on
 > Solaris, KCC on Linux, pgiCC on Linux, Visual C on Windows, CC on Compaq
 > Alpha and KCC on Solaris) AND when I changed the sequence of the includes
 > the error in the system header disappeared with all gcc versions and with
 > the Intel Compiler. So in my opinion from this investigation it's too easy
 > to say this is a user bug !
 >  
 > > Have you created the preprocessed source?  Can you send it to me?
 > > Can you send me the preprocessed source from a compiler that works?
 > 
 > Sure. See below ....
 
 No, that is not preprocessed source.  You sent me the include trace,
 which was *completely different for the two compilers*.  How can you
 expect to get the same result when the compilers are reading different
 header files?
 
 Neil.