Re: Galera FTBFS in f33, but same build passes in f32

[Jeff Law]

On Thu, 2020-08-27 at 14:21 +0200, Lukas Javorsky wrote:
>  Hi folks,
>  
>  I've run into the compilation problem in the Galera package
>  This problem occurs only on f33 and higher tho.
>  
>  Here is build performed on Fedora 32: 
> https://koji.fedoraproject.org/koji/taskinfo?taskID=50232504
>  
>  And here is the same build, but on Fedora 33: 
> https://koji.fedoraproject.org/koji/taskinfo?taskID=50232498
>  
>  Does anyone know what could cause this?
>  It looks like something that changed in cc1plus or scons is causing this, so if you had
> similar issue, please let me know. 
I'm pretty confident this is a source level
issue, not a problem with gcc or
scons.

If you look at the cpp output you'll find something like this:

# 130 "galera/tests/defaults_check.cpp" 3 4
       (
# 130 "galera/tests/defaults_check.cpp"
       it == map.end()
# 130 "galera/tests/defaults_check.cpp" 3 4
       ) ? _ck_assert_failed("galera/tests/defaults_check.cpp", 130,
"Failure '"
# 130 "galera/tests/defaults_check.cpp"
       "it == map.end()"
# 130 "galera/tests/defaults_check.cpp" 3 4
       "' occurred" ,
# 130 "galera/tests/defaults_check.cpp"
       "Failed to insert KV pair: %s = %s", param.first.c_str(),
param.second.c_str()
# 130 "galera/tests/defaults_check.cpp" 3 4
       , __null) : _mark_point("galera/tests/defaults_check.cpp", 130)

# 131 "galera/tests/defaults_check.cpp"

Where ck_assert_failed is defined by /usr/include/check.h, which comes from
check-devel:
CK_DLL_EXP void CK_EXPORT _ck_assert_failed(const char *file, int line,
                                            const char *expr, const char *msg,
                                            ...) CK_ATTRIBUTE_NORETURN
CK_ATTRIBUTE_FORMAT(printf, 4, 5);


Which says argument 4 of the call to ck_assert_failed is a printf style argument.
 

THat argument is:

"Failed to insert KV pair: %s = %s", param.first.c_str()

Note two string format arguments (%s).  Yet there are 3 arguments
(param.first.c_str(), param.second.c_str(), __null

That's what the compiler is complaining about.  This all indicates either check.h
is broken or the sources using it are broken.  But it isn't a compiler or scons
issue.


jeff