TNo, it is impossible to do properly without traits. You need a way to constrain the Key type to just hashable types, and that's just not implemented yet.
M@TeamPuzel Has anyone even made a proper hash function yet?
TI don't know, I doubt it
TYou could maybe implement a dictionary assuming the Key type can only be a string
MPriNova is working making one it seems.
TYeah that seems to be avoiding the problem by limiting it to strings keys
TCool, not sure what the
object thing does
MIts a type thats supposedly easier to use than
AnyTypeTOh, yeah, it looks like a runtime dynamic type feature
OThanks for your answers
SIIRC the object type is just a Variant over some internal structures, like
_RefCountedListRef and _ImmutableString. Definitely wouldn't recommend messing with them too much.
MI thought the
object type was supposed to be the nicer AnyType...SI'm not sure what AnyType actually is TBH. When Mojo spits MLIR errors about it, it seems to be of
!kgen.mlirtype which doesn't tell us all that much.MIs
object safe to use?SHmm, that's a hard question. I think as long as you're using the basic Bool and Float values it is, but I really don't know how to deal with the
_RefCountedListRef, _Function, and similar values. Probably best to stay away from them. (It also looks like they don't necessarily have attached / visible properties [or if they do have some there's very few], so messing with them at all might be hard. Maybe a couple MLIR ops are dedicated to them, I wouldn't know.)SJust for reference, this is what the LSP lists
o._value.value as (if o is of type object).MI just have so many issues come up with using
AnyType.
PThe interesting part is:
the underlying _ObjectImpl type, which can be used for conversion purposes and other stuff. The underlying type is the Variant type (MLIR type)
This seems similar to what is used as DodgyType in a community repo I saw.
The documentation misses a lot of functions, which can only be seen by the completion in a IDE.
the underlying _ObjectImpl type, which can be used for conversion purposes and other stuff. The underlying type is the Variant type (MLIR type)
This seems similar to what is used as DodgyType in a community repo I saw.
The documentation misses a lot of functions, which can only be seen by the completion in a IDE.
MI think that would be my repo with the DodgyString 
.
.PSo maybe you can substitute the Dodgy types with an 'object' as a test.
MThe DodgyType was due to Strings not being passable so a special type was needed to be created, though maybe I'm misunderstanding your statement.
POk, understand.
MCongrats @PriNova, you just advanced to level 10!
TAnyType looks a lot like the most generic of traits/protocols
TIts syntax also looks like a trait
TIt and Object have nothing in common, not sure where you heard that
TObject is a dynamic runtime object and AnyType is for constraining generics
I guess AnyType might feel a bit like boilerplate until you can also use other traits in its place
I guess AnyType might feel a bit like boilerplate until you can also use other traits in its place
PThe
!kgen seems to be a Fortran Kernel generator and is used for AST generation and manipulation.SYeah, the
kgen dialect seems to contain lots of special types for use in the AST or other. I was sort of surprised by the fact they moved !pop.pointer to !kgen.pointer, and there's probably more examples of types that might not belong. I'd guess the organization / categorization isn't perfect.SThere's also examples like
!kgen.declref that definitely make sense