On 10/08/2016 12:28 PM, Halil Pasic wrote: > > > On 10/08/2016 01:37 PM, Paolo Bonzini wrote: >>> Even though most put/get have no issues now, when somebody writes a new >>>> put, he or she could run into issues if only checking the type >>>> signature. It makes the code more readable. > >> No, it doesn't because one is left wondering what is VMS_LINKED about. > > I agree with Paolo. IMHO VMS_LINKED is conceptually difficult in a sense > that it's quite different that what we have currently. I have the feeling, > conceptually, we are trying to fit in something like data structures with a > type parameter (the element type) here. AFAIU what vmstate currently can is > directed acyclic graphs of certain stuff (and also completely custom > handling based on custom put/get). > You are right. What we have in vmstate now cannot handle a recursive or cyclic structure in a generic way. VMS_LINKED is intended to indicate a recursive structure, with the newly added parameters of put/get providing information about the element type in the recursive structure.
Thanks, Jianjun >> What is the relation between linked datastructures and passing some >> arguments as NULL/noon-NULL? > > IFAIU we need those for the datastructure because it's linked and has > a type parameter (element type). The two last arguments are for > the element type. These were added by the previous patch because > the old VMStateInfo did not need these. So we drastically changed > the scope of VMStateInfo with the previous patch and I'm not > sure I like this. > > I will have to stare at this a bit longer to bring something more > constructive than these (largely feeling-based) remarks. > > Cheers, > Halil >
