…t is sourced by bash_completion.
… overhead with big sum type the big sum language type have generated a lot of scalability problems (and fixes) and the benefit of the sum type was slight it remains the problem of the executable size (contains both full client and server code)
for some reason the client can ask a unknown client type during serialisation even wihtout having asking it previously on deserialisation. it may be that the type don't have to be explored completly during deserialsation (e.g. deserialisong option(toto) doesn't mean we will ask toto)
… of type definition types which are not statically used in client code or in server rpc code are now cleaned from cleint code remark: the client could still need such type at run-time, (e.g. polymorphic call back from server to client) when it happens, the client completes its type map by asking the server
…t with unknown type previously all type needed at run time should be on the client side now if a type is not known by the client, it can request its definition on the server this feature preprares more aggressive cleaning of the client code
…handle more case of generated code the detection is done on final code (optimised ...) which contains local anonymous applied function (as block, see fix on depth datastructures) and variable assignement with sharing, and while and switch ... the detection did not distinguish toplevel statement and in function statement, so it could not handle statement now we consider that local variable can be assign without side-effect, and handles block as applied anonymous function and we distinguish toplevel statement