Notes on talk with Niko about signals and types

parent 67bf674e
......@@ -183,3 +183,117 @@ typedef enum
} GSignalFlags;
* Discussion with Niko
<nmatsakis> ok so
<nmatsakis> you basically want to (for now?) hard code various names?
<federico> if we could cover the fundamental types (the definitions in ) for the first pass, it would be a *great* start
<federico> nmatsakis: yeah, I'm happy hardcoding for now
<nmatsakis> federico: did you understand my associated constant idea?
<nmatsakis> seems like that might be easier
<nmatsakis> that is, you might do something like this:
<nmatsakis> well, actually, maybe constant isn't what you want even
<federico> is that where a trait forces the impl to define some constants for its own case?
<nmatsakis> wait, are these GType constants specific to signals?
<nmatsakis> they're not, right?
<nmatsakis> (sorry, caching back in)
<nmatsakis> *paging
<federico> they are constants that gobject uses to identify types
<nmatsakis> ok, so, we have this trait and method:
<nmatsakis> you could imagine a base trait like
<federico> the fundamental types are really constants - that block in the URL above. What we define with our own gclass::get_type() are also GTypes, but those are generated at runtime
<nmatsakis> trait HasGType { fn get_type() -> GType; }
<nmatsakis> and then
<nmatsakis> impl HasGType for i32 { fn get_type() -> GType { G_TYPE_CHAR } }
<federico> niiiiiiice
<nmatsakis> and then you could just do `i32::get_type()`
<nmatsakis> well, probably this:
<nmatsakis> `<X as HasGType>::get_type()` where `X` is the "generated" version of the type the user wrote
<nmatsakis> (not that it matters, but after inlining etc the compiler will convert that into a constant anyway...)
<federico> nmatsakis: this is very elegant
<nmatsakis> the nice thing is, it will work automatically with the user's types...
<federico> nmatsakis: with that HasGType in place, would you remove get_type() from GInstance and instead impl HasGType for GInstance instead?
<nmatsakis> yep
<federico> awesome
<nmatsakis> probably `trait GInstance: HasGType`
<nmatsakis> and maybe some kind of more consistent naming scheme :P
<nmatsakis> (least of our worries I guess)
<federico> yep
<federico> on a related note:
<federico> while writing tests/, I wanted to do
<nmatsakis> PS I'm very excited to see you hacking on signals =)
<federico> let obj = Signaler::new();
<federico> let type = obj.get_type();
<federico> but that's not a method with a &self
<federico> so I did
<federico> let obj_type = Signaler::get_type();
<federico> but that's not really what I want; I'm calling a standalone get_type() function instead of a specific instance's
<nmatsakis> hmm
<federico> nmatsakis: I didn't quite grok how to move things around to do that
<nmatsakis> well yeah so you sort of want the ability to do both
<nmatsakis> (right?)
<federico> nmatsakis: I'm excited, too! I was rather scared of procedural macros, but I started reading grammar.lalrpop and went from there, and it's quite nice to work in
<federico> (... modulo deciphering error messages inside the proc macro)
<nmatsakis> in theory acrichto was working on fixing that
<nmatsakis> in particular, he was making it possible to have spans
<nmatsakis> I should ask him what's up with that
<federico> nmatsakis: yeah, I want a gimme_gtype_for_this_class_for_which_i_dont_have_an_instance() and gimme_gtype_for_this_instance(obj)
<nmatsakis> right so if we give them distinct names
<nmatsakis> would be easy enough
<nmatsakis> I guess have to look at where the best place would be
<nmatsakis> but it would look something like `fn get_dynamic_type(&self) -> GType`
<federico> right
<federico> probably get_instance_type() is more glib-like
<nmatsakis> k
<federico> ok, I'll look into that
<federico> having separate names makes it easier
<nmatsakis> seems like ..
<nmatsakis> probaly just add something to GInstance
<federico> so let's say I have the HasGType trait in place
<nmatsakis> e.g.
<nmatsakis> fn get_instance_type(&self) -> GType { some_function_that_gets_gtype_from_a_gobject(self.to_gobject_ptr()) }
<federico> then the generator would spit <sometype as HasGType>::get_type() - and if sometype is not supported, the compiler would flame me
<nmatsakis> yes
<nmatsakis> (unrelated, but one thing I've been contemplating, which would be relevant here,
<nmatsakis> is the ability to give more custom errors for cases like these)
<federico> hmm, I guess that's fine. I'd love to give back an error like "only these types are supported for gtyping: ..."
<nmatsakis> i.e., it'd be cool if we could make it say "this is not a valid signal type"
<nmatsakis> or something
<federico> yeah
<nmatsakis> yeah, for now that won't be as nice,
<nmatsakis> though there *is* a sort of hack here that is relevant
<nmatsakis> (it's unstable as all get out though)
<nmatsakis> (and not really on a path to stability)
<nmatsakis> which is that you can tag a trait with a message to display when it is not implemented
<nmatsakis> (but I want to do something much more flexible eventually)
<nmatsakis> still, might be useful to use for now
<federico> nmatsakis: in the future, G_TYPE_VARIANT is a thing that lets you define a complex type (struct with these fields, array of blah, etc.) - we can probably do that later
<nmatsakis> since we're using unstable features a lot anyway :)
<nmatsakis> ah, yeah
<federico> ok, I'll try to get this going tomorrow
<federico> nmatsakis: this is really useful info, thank you :)
<federico> what else did I want to ask...
<nmatsakis> federico: my pleasure, will of course be around tomorrow too :)
<federico> nmatsakis: just so I know - if the grammar reads in, say "i32", what Type would ast get for that?
<nmatsakis> federico:
<nmatsakis> I believe
<nmatsakis> then probably
<nmatsakis> is the Path
<nmatsakis> with
<nmatsakis> being `i32`
<nmatsakis> (or the interned version thereof)
<federico> aha
<federico> what's PathId.tys?
<federico> uuh, if I impl HasGType for i32 and stuff, can that go into gnome-class-shims? what with it being an extern crate... or does that impl need to go in the generated code?
<federico> ok, gk
<federico> going to the supermarket
<federico> nmatsakis: thanks again; I now have lots of interesting stuff to do tomorrow :)
<nmatsakis> federico: nice
<nmatsakis> federico: ah, I missed your question about PathId.tys --
<nmatsakis> federico: I think that refers to a path like `Vec<u32, f32>`
<nmatsakis> federico: the u32, f32 are the "tys"
<nmatsakis> federico: as for the impls of HAsGType, they would go in the same crate that defines the trait
<nmatsakis> federico: so gnome-class-shims
<nmatsakis> federico: in particular I mean the impls for "builtin" types like u32 etc
<federico> nmatsakis: got it, thanks
Markdown is supported
0% or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment