GIMPLE and datastructures

[Jan Hubicka]

Hi,
the PR 13405 is related to bugreport where BIND_EXPR contains
initializer with non-GIMPLE PTRMEM_CST expression.  What is our strategy
when dealing with data initializers?  Do we want to lower them into
something language independent, or do we want language dependencies to
leak into GIMPLE this way?

In the first case I will re-add analyze_expr hook to deal with this
correctly, but it would really simplify my life if we were able to avoid
them.  How hard is to gimplify this into ADDR_EXPR?

Honza