#303 Segfault during static initialization

[Ray Logel]

Encountered a segfault during program start, it occured in loglevel.cxx:

static
LogLevel
defaultStringToLogLevelMethod(const tstring& s)
{
    std::size_t const tbl_size
        = sizeof (log_levels_table) / sizeof (log_levels_table[0]);

    for (log_levels_table_rec const * it = log_levels_table;
        it != log_levels_table + tbl_size; ++it)
    {
        if (*it->str == s)  /////// Segfaulting Here ///////
            return it->ll;
    }

    return NOT_SET_LOG_LEVEL;
}

The problem was that elements in the static log_levels_table hadn't been constructed yet.

The problem is well know in C++, this website explains it pretty well.
http://www.parashift.com/c++-faq/static-init-order.html

[Václav Haisman]

Well, there is nothing much I can do about it. You simply cannot use log4cplus before entering main() at least.

[Marcel Loose]

That's unacceptable. You do not always have control over when the first log message is being generated. It can very well be produced by a statically initialized object.

[Václav Haisman]

What do you want me to do about it? I ain't no magician.

[Marcel Loose]

True ;)
Looking at the code, I doubt it's a static initialization order fiasco.
The variable log_levels_table is defined in the same compilation unit as defaultStringToLogLevelMethod(). I must be missing something.

[Václav Haisman]

Are you using log4cplus as static library? I think it still could be static initialization order fiasco if you were using log4cplus as a static library.

[Václav Haisman]

You might get away with calling log4cplus::initialize() yourself before invoking log4cplus from your static initializers.

[Ray Logel]

We are using a global static instance of log4cplus::Logger in most of our cpp files. During this construction it attempts to use defaultStringToLogLevelMethod some where which isn't initialized yet.

I have attached possible solution.

[Václav Haisman]

What compiler are you using?

[Ray Logel]

g++ (GCC) 4.4.7 20120313 (Red Hat 4.4.7-4) on RedHat Enterprise 6.5

[Václav Haisman]

I have your patch with some additional changes to 1.2.x branch. Please test it.

[Ray Logel]

Thanks, we will try and test it.

[Václav Haisman]

And I have pushed similar patches to 1.1.x and master as well.

[Václav Haisman]

I had to revert the change because it made static log4cplus library non-functional. I am afraid we are back to "do not use log4cplus before main()" situation.

[Václav Haisman]

The "do not use log4cplus before main()" advice remains. I am closing this.