2010/3/8 Ian Lance Taylor <iant@xxxxxxxxxx>: > Juan Carlos Franzoy <jfranzoy@xxxxxxxxx> writes: ... > > The backtrace in http://gcc.gnu.org/ml/gcc-help/2010-03/msg00090.html > does not make much sense to me. It makes less sense to me. > ... Somehow calling > ats::base::getEnvironmentVariable leads to a call to _init. How does > that happen? Do you have different shared libraries involved? Yes, actually there are 3 shared libraries involved. 1) libAtsBase: The one wich has Loggers 2) libAtsStore: wich has the first statically Logger object constructed (frame #19) 3) libRTDB: which has the second Logger object statically and reentrantly constructed. The funny fact is that de dependencies between them are: libAtsStore -> libRTDB ->libAtsBase libAtsStore -> libAtsBase The arrows meands 'use'. I would espect that something in libRtdb could trigger libAtsBase static initialization, not the reverse. > ... It seems that perhaps a symbol in one shared library is referring to a > symbol defined in a different shared library, and that the Solaris > dynamic linker is deciding to run the static initializers in the > latter library. This is then leading to the reentrant call. > > Ian > Anyway. I will investigate it later. Until that I will suppouse "Any code run in static initialization that call code on another translation unit may trigger static initialization of other objects". Thanks, very much Ian for your help.
