I'm proposing the addition of a new types of map/list/set with specified child types. There is no doubt about the need of a map which keys are objects (like in groupCount()) and the benefits of using a g:Map over plain json objects in GraphSON3.
> is that basically the core of the usability problem? I think its a specification problem, not a usability problem. GraphTraversal interface exposes valueMap() as a method that returns a Map<String, E2>, but the GraphSON3 specification doesn't provide an specification of a Map with string keys so it's being returned as an Map<Object, Object>. > Is there a way to do that? how do we know what's in the Map/List until we iterate it all? I think its a responsibility of the graph processor, that could optimize by returning a specific type of traverser as a result of the execution of a certain traversal (in the example, when the valueMap is the last step). On Tue, Sep 19, 2017 at 6:56 PM, Stephen Mallette <[email protected]> wrote: > > user is stuck with a unspecified map, which contains strings as keys. > > so basically in .NET the result of valueMap() is stuck with "object" as a > key when the user expects "string". i guess that's inconvenient. if you > iterated the keys of the dictionary and wanted to do some string operation > on that you'd have to cast - is that basically the core of the usability > problem? i tend to think it's more inconvenient to not be able to get a > proper result from groupCount() where the keys might be something other > than a string - we definitely don't want to lose that capability. > > > tackle this at the source and try to specify the child types on > serialization when we have that information. > > Is there a way to do that? how do we know what's in the Map/List until we > iterate it all? > > On Tue, Sep 19, 2017 at 12:38 PM, Jorge Bay Gondra < > [email protected] > > wrote: > > > I wanted to bump this open discussion that after the JIRA adjustments > > probably got buried. > > > > On Tue, Sep 12, 2017 at 2:31 PM, Jorge Bay Gondra < > > [email protected]> > > wrote: > > > > > Collections with unspecified child types are useful for mixed types, > but > > > we are reusing those types of collections for all the cases. > > > This behaviour poses a problem for strict statically typed languages > like > > > C#. Let's see if I can explain myself with an example: > > > > > > valueMap() step yields a map representation. In GraphSON2, it was > > returned > > > as a js object (where keys are always strings), that were mapped to C# > > > Dictionary<string, object> (equivalent of java's Map<String, Object>). > > > In GraphSON3, it's returned as a "g:Map", that is mapped to C# > > > Dictionary<object, object>. As there is no type coercion (ie: type > > erasure > > > in java) for Dictionary<string, object> to Dictionary<object, object>, > > the > > > user is stuck with a unspecified map, which contains strings as keys. > > > > > > This causes a problem in any statically typed languages (outside java > > > generics w/ type-erasure thingy). > > > > > > There are workarounds (like exposing conversion methods or lazy > > > deserialization) but maybe we can tackle this at the source and try to > > > specify the child types on serialization when we have that information. > > > > > > On Mon, Sep 11, 2017 at 6:04 PM, Stephen Mallette < > [email protected]> > > > wrote: > > > > > >> Mixed types would be the problem - that's why objects within the > > List/Map > > >> have types embedded within them. > > >> > > >> On Mon, Sep 11, 2017 at 12:01 PM, Robert Dale <[email protected]> > > wrote: > > >> > > >> > I'm not sure that's feasible since any list or map could contain > mixed > > >> > types. Maybe I don't understand the use case. > > >> > > > >> > Robert Dale > > >> > > > >> > On Mon, Sep 11, 2017 at 11:27 AM, Jorge Bay Gondra < > > >> > [email protected] > > >> > > wrote: > > >> > > > >> > > Hi, > > >> > > The new GraphSON3 g:List, g:Set and g:Map definitions are an > > >> improvement > > >> > > over js arrays and associative arrays, but they doesn't provide > the > > >> child > > >> > > type information. > > >> > > > > >> > > We could add support for chid type info, something like > > >> > > "g:List<g:Traverser>" or "g:Map<g:Int32,g:Vertex>". > > >> > > > > >> > > For typed languages, it would allow compile time checks. > > >> > > For any technology, it would more user-friendly error messages > > >> ("Expected > > >> > > Vertex, obtained..."). > > >> > > > > >> > > Jorge > > >> > > > > >> > > > >> > > > > > > > > >
