[cfe-commits] [Patch] -Wformat: properly handle length modifiers used with %n (Was: -Wformat: warn about using length modifiers with %n)

[Hans Wennborg]

On Tue, Aug 7, 2012 at 6:12 PM, Peter Cooper <peter_cooper at apple.com> wrote:
> Hi Hans,
>
> This series of commits seems to be breaking the gcc test suite.  Could you please investigate?
>
> The output we have is

> gcc.dg/format/c90-printf-1.c   bad argument types (test for warnings, line 200)
> gcc.dg/format/c90-printf-1.c  -DWIDE  bad argument types (test for warnings, line 200)

This is: printf ("%n", un);

Where 'un' is 'unsigned int *'. GCC warns in -pedantic mode that it
expects 'int *', but Clang doesn't warn since it doesn't do anything
special with -pedantic for this kind of checks.

> gcc.dg/format/c99-printf-1.c   %hhn plain char (test for warnings, line 195)
> gcc.dg/format/c99-printf-1.c   %hhn unsigned char (test for warnings, line 196)
> gcc.dg/format/c99-printf-1.c  -DWIDE  %hhn plain char (test for warnings, line 195)
> gcc.dg/format/c99-printf-1.c  -DWIDE  %hhn unsigned char (test for warnings, line 196)

Same thing here: gcc in -pedantic mode warns that it expects 'signed
char *', not 'char *' or 'unsigned char *'. Clang again doesn't do
anything special with -pedantic.


The reason some of these tests passed before was that since my
r160966, Clang would warn about any use of %n where the argument
wasn't exactly 'int*'. This was wrong, and broke a lot of other tests
in the suite, so the current results are a net win :)

We could make -Wformat do strict argument checking with -pedantic, but
I don't know how important that is?

Thanks,
Hans