1 | Proposal For Use of Virtual Tables |
---|
2 | ================================== |
---|
3 | |
---|
4 | This is an adaptation of the earlier virtual proposal, updating it with new |
---|
5 | ideas, reframing it and laying out more design decisions. |
---|
6 | |
---|
7 | The basic concept of a virtual table (vtable) is the same here as in most |
---|
8 | other languages. They will mostly contain function pointers although they |
---|
9 | should be able to store anything that goes into a trait. |
---|
10 | |
---|
11 | This should replace the virtual proposal, although all not all features have |
---|
12 | been converted to the new design. |
---|
13 | |
---|
14 | Trait Instances |
---|
15 | --------------- |
---|
16 | |
---|
17 | Currently traits are completely abstract. Data types might implement a trait |
---|
18 | but traits are not themselves data types. This will change that and allow |
---|
19 | instances of traits to be created from instances of data types that implement |
---|
20 | the trait. |
---|
21 | |
---|
22 | trait combiner(otype T) { |
---|
23 | void combine(T&, int); |
---|
24 | }; |
---|
25 | |
---|
26 | struct summation { |
---|
27 | int sum; |
---|
28 | }; |
---|
29 | |
---|
30 | void ?{}( struct summation & this ) { |
---|
31 | this.sum = 0; |
---|
32 | } |
---|
33 | |
---|
34 | void combine( struct summation & this, int num ) { |
---|
35 | this.sum = this.sum + num; |
---|
36 | } |
---|
37 | |
---|
38 | trait combiner obj = struct summation{}; |
---|
39 | combine(obj, 5); |
---|
40 | |
---|
41 | Internally a trait object is a pair of pointers. One to an underlying object |
---|
42 | and the other to the vtable. All calls on an trait are implemented by looking |
---|
43 | up the matching function pointer and passing the underlying object and the |
---|
44 | remaining arguments to it. |
---|
45 | |
---|
46 | Trait objects can be copied and moved by copying and moving the pointers. |
---|
47 | They should also be able to own or borrow the underlying object. |
---|
48 | |
---|
49 | Hierarchy |
---|
50 | --------- |
---|
51 | |
---|
52 | Virtual tables by them selves are not quite enough to implement the planned |
---|
53 | hierarchy system. An addition of type ids, implemented as pointers which |
---|
54 | point to your parent's type id, is required to actually create the shape of |
---|
55 | the hierarchy. However vtables would allow behaviour to be carried with the |
---|
56 | tree. |
---|
57 | |
---|
58 | The hierachy would be a tree of types, of traits and structs. Currently we do |
---|
59 | not support structural extension, so traits form the internal nodes and |
---|
60 | structures the leaf nodes. |
---|
61 | |
---|
62 | The syntax is undecided but it will include a clause like `virtual (PARENT)` |
---|
63 | on trait and struct definitions. It marks out all types in a hierarchy. |
---|
64 | PARENT may be omitted, if it is this type is the root of a hierachy. Otherwise |
---|
65 | it is the name of the type that is this type's parent in the hierarchy. |
---|
66 | |
---|
67 | Traits define a trait instance type that implements all assertions in this |
---|
68 | trait and its parents up until the root of the hierarchy. Each trait then |
---|
69 | defines a vtable type. Structures will also have a vtable type but it should |
---|
70 | be the same as their parent's. |
---|
71 | |
---|
72 | Trait objects within the tree can be statically cast to a parent type. Casts |
---|
73 | from a parent type to a child type are conditional, they check to make sure |
---|
74 | the underlying instance is an instance of the child type, or an instance of |
---|
75 | one of its children. The type then is recoverable at runtime. |
---|
76 | |
---|
77 | As with regular trait objects, calling a function on a trait object will cause |
---|
78 | a lookup on the the virtual table. The casting rules make sure anything that |
---|
79 | can be cast to a trait type will have all the function implementations for |
---|
80 | that trait. |
---|
81 | |
---|
82 | Converting from a concrete type (structures at the edge of the hierchy) to |
---|
83 | an abstract type works the same as with normal trait objects, the underlying |
---|
84 | object is packaged with a virtal table pointer. Converting back to an abstract |
---|
85 | type requires confirming the underlying type matches, but then simply extracts |
---|
86 | the pointer to it. |
---|
87 | |
---|
88 | ### Inline vtables |
---|
89 | Since the structures here are usually made to be turned into trait objects |
---|
90 | it might be worth it to have fields on them to store the virtual table |
---|
91 | pointer. This would have to be declared on the trait as an assertion, but if |
---|
92 | it is the trait object could be a single pointer. |
---|
93 | |
---|
94 | It is trival to do if the field with the virtual table pointer is fixed. |
---|
95 | Otherwise some trickery with pointing to the field and storing the offset in |
---|
96 | the virtual table to recover the main object would have to be used. |
---|
97 | |
---|
98 | ### Virtual Tables as Types |
---|
99 | Here we consider encoding plus the implementation of functions on it. Which |
---|
100 | is to say in the type hierarchy structures aren't concrete types anymore, |
---|
101 | instead they are parent types to vtables, which combine the encoding and |
---|
102 | implementation. |
---|
103 | |
---|
104 | Resolution Scope |
---|
105 | ---------------- |
---|
106 | |
---|
107 | What is the scope of a resolution? When are the functions in a vtable decided |
---|
108 | and how broadly is this applied? |
---|
109 | |
---|
110 | ### Type Level: |
---|
111 | Each structure has a single resolution for all of the functions in the |
---|
112 | virtual trait. This is how many languages that implement this or similar |
---|
113 | features do it. |
---|
114 | |
---|
115 | The main thing CFA would need to do it this way is some single point where |
---|
116 | the type declaration, including the functions that satify the trait, are |
---|
117 | all defined. Currently there are many points where this can happen, not all |
---|
118 | of them will have the same definitions and no way to select one over the |
---|
119 | other. |
---|
120 | |
---|
121 | Some syntax would have to be added. All resolutions can be found at compile |
---|
122 | time and a single vtable created for each type at compilation time. |
---|
123 | |
---|
124 | ### Explicate Resolution Points: |
---|
125 | Slightly looser than the above, there are explicate points where the vtables |
---|
126 | are resolved, but there is no limit on the number of resolution points that |
---|
127 | might be provided. Each time a object is bound to a trait, one of the |
---|
128 | resolutions is selected. This might be the most flexable option. |
---|
129 | |
---|
130 | An syntax would have to be provided as above. There may also be the option |
---|
131 | to name resolution points so that you can choose between them. This also |
---|
132 | could come with the ablity to forward declare them. |
---|
133 | |
---|
134 | Especially if they are not named, these resolution points should be able to |
---|
135 | appear in functions, where the scoping rules can be used to select one. |
---|
136 | However this also means that stack-allocated functions can end up in the |
---|
137 | vtable. |
---|
138 | |
---|
139 | ### Site Based Resolution: |
---|
140 | Every place in code where the binding of a vtable to an object occurs has |
---|
141 | its own resolution. Syntax-wise this is the simplest as it should be able |
---|
142 | to use just the existing declarations and the conversion to trait object. |
---|
143 | It also is very close to the current polymorphic reolution rules. |
---|
144 | |
---|
145 | This works as the explicate resolution points except the resolution points |
---|
146 | are implicate and their would be no selection of which resolution to use. The |
---|
147 | closest (current) resolution is always selected. |
---|
148 | |
---|
149 | This could easily lead to an explosition of vtables as it has the most fine |
---|
150 | grained resolution the number of bindings in a single scope (that produces |
---|
151 | the same binding) could be quite high. Merging identical vtables might help |
---|
152 | reduce that. |
---|
153 | |
---|
154 | Vtable Lifetime Issues |
---|
155 | ---------------------- |
---|
156 | |
---|
157 | Vtables interact badly with the thunk issue. Conceptually vtables are static |
---|
158 | like type/function data they carry, as those decitions are made by the |
---|
159 | resolver at compile time. |
---|
160 | |
---|
161 | Stack allocated functions interact badly with this because they are not |
---|
162 | static. There are serveral ways to try to resolve this, however without a |
---|
163 | general solution most can only buy time. |
---|
164 | |
---|
165 | Filling in some fields of a static vtable could cause issues on a recursive |
---|
166 | call. And then we are still limited by the lifetime of the stack functions, as |
---|
167 | the vtable with stale pointers is still a problem. |
---|
168 | |
---|
169 | Dynamically allocated vtables introduces memory management overhead and |
---|
170 | requires some way to differentate between dynamic and statically allocated |
---|
171 | tables. The stale function pointer problem continues unless those becomes |
---|
172 | dynamically allocated as well which gives us the same costs again. |
---|
173 | |
---|
174 | Stack allocating the vtable seems like the best issue. The vtable's lifetime |
---|
175 | is now the limiting factor but it should be effectively the same as the |
---|
176 | shortest lifetime of a function assigned to it. However this still limits the |
---|
177 | lifetime "implicately" and returns to the original problem with thunks. |
---|