sizeof size_t != sizeof unsigned long

[Dirk Mueller]

On Sunday 22 February 2004 20:16, Nicolas Goutte wrote:

> So we are talking about gcc 2.96, not about kcc or egcs 1.2 (or whatever to
> call it.)

Sure. as I said, gcc 2.96 does not support the warning option. However, the 
g++ compiler (which we use to base the test on), was not 2.96: 

configure:3597: checking for C++ compiler version
configure:3600: g++ --version </dev/null >&5
g++ (GCC) 3.3.2


Thats what I tried to explain you before: We test the warnings with the C++ 
compiler, but then from time to time use the C compiler. The test with the 
C++ compiler passes, since its a reasonably new one. The C compiler is a 
broken old 2.96, so it will fail in funky ways. 

The same thing can happen the other way around: we could find a g++ that 
points to 2.96, and a cc that points to 3.3.2. 

I see the following solutions: 

a) disable warnings by default. Makes most sense I guess. 

b) run the test for the -Wformat flags twice, once for the c compiler, and 
once for the C++ compiler. Slow. 

c) change the error message to tell them that they have a broken setup and 
that they should fix it. 

d) tell those users that run into the problem to configure with 
--disable-warnings.

e) only add the Wformat* flags to the CXXFLAGS, and don't set them for the C 
compiler. Is a security problem, since we have some .c based sources in CVS. 

> Also RH means North-American market, India etc. Do you want to lose these
> markets by not trying to support gcc 2.96?

Boy, stop talking on that level if you can't even get the facts straight. It 
does not happen on a plain RH 7.3 installation. Please understand that. 

It only happens on systems where the user misconfigured their system. There 
are plenty ways of breaking your system, we can't check for all. Lots of 
those end up in a crashing dcopserver or similar obvious errors. 

Furthermore, those redhat distros using gcc 2.96 are > 2 years old, their 
maintenance period expired long ago. Anybody who is still using such software 
makes a mistake anyway. You can't expect that you build the new mercedes 
engine into your 20 year old fiat panda, do you?

Please review the patch which should solve the issue. I think it makes most 
sense for the branch, but I think its right for HEAD too, since we'll forget 
to disable warnings for the upcoming KDE 3.3 release then again. 


Dirk
-------------- next part --------------
A non-text attachment was scrubbed...
Name: acinclude.m4.in.diff
Type: text/x-diff
Size: 981 bytes
Desc: not available
URL: <http://mail.kde.org/pipermail/kde-core-devel/attachments/20040222/d2946f66/attachment.diff>