On 03/17/2009 03:26 AM, Henrik Nordstrom wrote: > The more I think on this the more I am of the opinion that using > self-registering static/global objects as method of initialization & > registration is generally a mistake. Better if each such class have a > method for initialization, with initialization order explicitly coded in > the main program. Also makes transition to runtime loaded modules easier > and less intrusive as each module can assume the modules it registers > into has been properly initialized already which means it can do a full > initialization.
I do not want to make any general statements, but the ESI layout patch I have just posted implements explicit registration of parsers, initiated by the Esi::Init() call in main.cc. If somebody is willing to do the equivalent for the ACL code, please post a patch. The ACL changes are just a little more complex than the posted ESI changes and the code to be massaged is already isolated in AclRegs.cc. Meanwhile, I will focus on more pressing issues. Thank you, Alex.
