On Mon, 20 May 2019 at 04:07, Wx L <lwxiang1994@xxxxxxxxx> wrote:
>
> hi, I used gcc5.5.0(configure only with --prefix), glibc2.17 to compile
> some code with *conversion-null *warning in it.
> it is every simple so I put it here:
>
> #include <stdio.h>
> int main()
> {
> return NULL;
> }
>
> *What I Got*
> I did compile with:
> g++ -Wall -S test.c
> as expected, here came out the warning "warning: converting to non-pointer
> type 'int' from NULL [-Wconversion-null]"
>
> then I did compile step by step:
> g++ -Wall -E test.c
> g++ -Wall -S test.i
>
> the warning just disappeared, nothing came out.
>
> *What I have Tried*
> I tried to figure out if all warning will be disabled, so I compile some
> code with different warning in it, just *conversion-null *will be disabled
> when separate-step compiling.
> I tried to figure out if all gcc version behave the same way, and I found
> gcc5.5.0,gcc7.1.0,gcc7.3.0 did, but gcc4.8.5 works when separate-step
> compiling.
> After comparing the preprocessing files(*.i) from gcc4.8.5 and gcc5.5.0, I
> found that gcc5.5.0 put some linemarkers in it while gcc4.8.5 don't.
> here is the line markers:
>
> int main()
> {
> int a =
> # 4 "a.c" 3 4
> __null
> # 4 "a.c"
> ;
> return a;
> }
>
> if I do preprocess with gcc4.8.5, and do compile with gcc5.5.0, it warns
> well.
> if I do preprocess with gcc5.5.0, and do compile with gcc4.8.5, it can not
> warn *conversion-null.*
> if I do preprocess with gcc5.5.0 and add -P option(remove linemarkers), and
> do compile with gcc5.5.0, it warns well.
>
> Now I am sure that it is the linemarkers added by gcc5.5.0 preprocessor
> that cause the *conversion-null *warning not work as expected.
> But why one-step compile works? So I add -save-temps to one-step compile to
> see if it got a different preprocessing file.
> just like:
>
> g++ -Wall -save-temps -S test.c
>
> and now it can not warns too! the preprocessing file is exactly same as the
> one with "g++ -Wall -E test.c"
>
> *What Help I Need*
> After all the test, I confuse that if there is a bug with warning analyzer
> when analyzing the *conversion-null *warning code which around by
> linemarkers?
I think the problem is that the definition of NULL comes from a system
header, and GCC gets confused when issuing diagnostics related to
macros defined in system headers.
If you use -Wsystem-headers then you get a -Wconversion-null warning
even when preprocessing and compiling separately. This is related to
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=77513
Please report a bug.
> Is that -P a workaround to solve this problem when separate-step compiling?
> Will -P leads to some other problems?
Yes, removing all line directives means the compiler doesn't know
which code comes from system headers, and so some standard library
features will no longer compile cleanly. The -P flag should be used
when the output of the preprocessor is not going to be consumed by the
compiler, it shouldn't be used for C or C++ code.
