import statement location effects on random stability

[ejessen]

Hi,

I'm trying to determine if "import" statements should be within package statements or not, when it comes to random stability.

[uwes]

"import" just makes the symbols of the package visible. doing so inside or outside of a package can change what is being included/referenced and might change initialization order and therefore randomization. in general you should

 

- only import within a scope (such as in a package/module/program) and not use the global scope

- avoid (static) initialization races

- protect areas which you want rand-stable with code (or use uvm for that)

[dave_59]

You must import a package or explicitly reference a symbol in a package if you want to use it. Importing a symbol just means you don't have to prefix the symbol with the package name (i.e. pkg_name::symbol_name)

 

The biggest problems we see with packages related to random stability are caused by static variable initializations that call class constructors. You either have a class variable in your package that is statically initialized, or there is a static member of a class that is statically initialized. The ordering of these initializations is not deterministic. (http://www.parashift.com/c++-faq/static-init-order.html). What makes it even worse is the unanswered question: "if I compile my a package and nobody references it, does it exist?" meaning do the static variables exist and call functions/constructors as part of their initialization. Tools have different ways of handling this.