More systemantic treatment of bindings

Introduce `Bindings` and `BindingInfo` structs, as better system and
naming to replace the `var_info` dict and `next_var_id` ref.

Introduce a new kind `K"BindingId"` - an integer handle for an entity
(variable, constant, etc). Normally these are generated from `NameKey`
(`name,scope_layer`) pairs written in the user's program. But they can
also be generated programmatically by the `ssavar()` function, or by
`new_mutable_var()`. This should allow proper "gensym" bindings from
macros, without needing to rely on the generation of unique names via
name mangling.

In this system, `SSAValue` is just a particular kind of binding,
relegated to a low level IR concept which has invariants which can, in
principle, be checked or inferred by the variable analysis pass.

src/ast.jl

@@ -2,22 +2,61 @@
 abstract type AbstractLoweringContext end
 
 """
-Unique symbolic identity for a variable
+Unique symbolic identity for a variable, constant, label, or other entity
 """
-const VarId = Int
+const IdTag = Int
+
+"""
+Metadata about a binding
+"""
+struct BindingInfo
+    name::String
+    mod::Union{Nothing,Module} # Set when `kind === :global`
+    kind::Symbol              # :local :global :argument :static_parameter
+    is_ssa::Bool              # Single assignment, defined before use
+    is_ambiguous_local::Bool  # Local, but would be global in soft scope (ie, the REPL)
+end
+
+"""
+Metadata about "entities" (variables, constants, etc) in the program. Each
+entity is associated to a unique integer id, the BindingId. A binding will be
+inferred for each *name* in the user's source program by symbolic analysis of
+the source.
+
+However, bindings can also be introduced programmatically during lowering or
+macro expansion: the primary key for bindings is the `BindingId` integer, not
+a name.
+"""
+struct Bindings
+    info::Vector{BindingInfo}
+end
+
+Bindings() = Bindings(Vector{BindingInfo}())
+
+function new_binding(bindings::Bindings, info::BindingInfo)
+    push!(bindings.info, info)
+    return length(bindings.info)
+end
+
+function lookup_binding(bindings::Bindings, id::Integer)
+    bindings.info[id]
+end
+
+function lookup_binding(bindings::Bindings, ex::SyntaxTree)
+    # TODO: @assert kind(ex) == K"BindingId"
+    bindings.info[ex.var_id]
+end
+
+function lookup_binding(ctx::AbstractLoweringContext, id)
+    lookup_binding(ctx.bindings, id)
+end
 
 const LayerId = Int
 
 function syntax_graph(ctx::AbstractLoweringContext)
     ctx.graph
 end
 
-function new_var_id(ctx::AbstractLoweringContext)
-    id = ctx.next_var_id[]
-    ctx.next_var_id[] += 1
-    return id
-end
-
 #-------------------------------------------------------------------------------
 # AST creation utilities
 _node_id(ex::NodeId) = ex
@@ -67,7 +106,7 @@ function makeleaf(ctx, srcref, k::Kind, value; kws...)
     graph = syntax_graph(ctx)
     if k == K"Identifier" || k == K"core" || k == K"top" || k == K"Symbol" || k == K"globalref"
         makeleaf(graph, srcref, k; name_val=value, kws...)
-    elseif k == K"SSAValue"
+    elseif k == K"BindingId"
         makeleaf(graph, srcref, k; var_id=value, kws...)
     elseif k == K"label"
         makeleaf(graph, srcref, k; id=value, kws...)
@@ -104,9 +143,21 @@ unused(ctx, ex) = core_ref(ctx, ex, "UNUSED")
 
 top_ref(ctx, ex, name) = makeleaf(ctx, ex, K"top", name)
 
-# Create a new SSA variable
+# Create a new SSA binding
 function ssavar(ctx::AbstractLoweringContext, srcref)
-    makeleaf(ctx, srcref, K"SSAValue", var_id=new_var_id(ctx))
+    # TODO: Store this name in only one place? Probably use the provenance chain?
+    name = "ssa"
+    id = new_binding(ctx.bindings, BindingInfo(name, nothing, :local, true, false))
+    # Create an identifier
+    nameref = makeleaf(ctx, srcref, K"Identifier", name_val=name)
+    makeleaf(ctx, nameref, K"BindingId", var_id=id)
+end
+
+# Create a new local mutable variable
+function new_mutable_var(ctx::AbstractLoweringContext, srcref, name)
+    id = new_binding(ctx, BindingInfo(name, nothing, :local, false, false))
+    nameref = makeleaf(ctx, srcref, K"Identifier", name_val=name)
+    makeleaf(ctx, nameref, K"BindingId", var_id=id)
 end
 
 # Assign `ex` to an SSA variable.

src/desugaring.jl

@@ -10,7 +10,7 @@ end
 
 struct DesugaringContext{GraphType} <: AbstractLoweringContext
     graph::GraphType
-    next_var_id::Ref{VarId}
+    bindings::Bindings
     scope_layers::Vector{ScopeLayer}
     mod::Module
 end
@@ -21,9 +21,9 @@ function DesugaringContext(ctx)
                               source=SourceAttrType,
                               value=Any, name_val=String,
                               scope_type=Symbol, # :hard or :soft
-                              var_id=VarId,
+                              var_id=IdTag,
                               lambda_info=LambdaInfo)
-    DesugaringContext(graph, ctx.next_var_id, ctx.scope_layers, ctx.current_layer.mod)
+    DesugaringContext(graph, ctx.bindings, ctx.scope_layers, ctx.current_layer.mod)
 end
 
 # Flatten nested && or || nodes and expand their children
@@ -691,7 +691,7 @@ function expand_forms_2(ctx::DesugaringContext, ex::SyntaxTree, docs=nothing)
     else
         if k == K"="
             @chk numchildren(ex) == 2
-            if kind(ex[1]) ∉ KSet"Identifier Placeholder SSAValue"
+            if kind(ex[1]) ∉ KSet"Identifier Placeholder BindingId"
                 TODO(ex, "destructuring assignment")
             end
         end

src/eval.jl

@@ -98,9 +98,9 @@ end
 
 # Convert SyntaxTree to the CodeInfo+Expr data stuctures understood by the
 # Julia runtime
-function to_code_info(ex, mod, funcname, nargs, var_info, slot_rewrites)
+function to_code_info(ex, mod, funcname, nargs, bindings, slot_rewrites)
     input_code = children(ex)
-    code = Any[to_lowered_expr(mod, var_info, ex) for ex in input_code]
+    code = Any[to_lowered_expr(mod, bindings, ex) for ex in input_code]
 
     debuginfo = ir_debug_info(ex)
 
@@ -115,7 +115,7 @@ function to_code_info(ex, mod, funcname, nargs, var_info, slot_rewrites)
     slot_rename_inds = Dict{String,Int}()
     slotflags = Vector{UInt8}(undef, nslots)
     for (id,i) in slot_rewrites
-        info = var_info[id]
+        info = lookup_binding(bindings, id)
         name = info.name
         ni = get(slot_rename_inds, name, 0)
         slot_rename_inds[name] = ni + 1
@@ -176,7 +176,7 @@ function to_code_info(ex, mod, funcname, nargs, var_info, slot_rewrites)
     )
 end
 
-function to_lowered_expr(mod, var_info, ex)
+function to_lowered_expr(mod, bindings, ex)
     k = kind(ex)
     if is_literal(k) || k == K"Bool"
         ex.value
@@ -203,7 +203,7 @@ function to_lowered_expr(mod, var_info, ex)
     elseif k == K"SSAValue"
         Core.SSAValue(ex.var_id)
     elseif k == K"return"
-        Core.ReturnNode(to_lowered_expr(mod, var_info, ex[1]))
+        Core.ReturnNode(to_lowered_expr(mod, bindings, ex[1]))
     elseif is_quoted(k)
         if k == K"inert"
             ex[1]
@@ -215,7 +215,7 @@ function to_lowered_expr(mod, var_info, ex)
             "top-level scope" :
             "none"              # FIXME
         nargs = length(ex.lambda_info.args)
-        ir = to_code_info(ex[1], mod, funcname, nargs, var_info, ex.slot_rewrites)
+        ir = to_code_info(ex[1], mod, funcname, nargs, bindings, ex.slot_rewrites)
         if ex.lambda_info.is_toplevel_thunk
             Expr(:thunk, ir)
         else
@@ -226,7 +226,7 @@ function to_lowered_expr(mod, var_info, ex)
     elseif k == K"goto"
         Core.GotoNode(ex[1].id)
     elseif k == K"gotoifnot"
-        Core.GotoIfNot(to_lowered_expr(mod, var_info, ex[1]), ex[2].id)
+        Core.GotoIfNot(to_lowered_expr(mod, bindings, ex[1]), ex[2].id)
     else
         # Allowed forms according to https://docs.julialang.org/en/v1/devdocs/ast/
         #
@@ -243,7 +243,7 @@ function to_lowered_expr(mod, var_info, ex)
         if isnothing(head)
             TODO(ex, "Unhandled form for kind $k")
         end
-        Expr(head, map(e->to_lowered_expr(mod, var_info, e), children(ex))...)
+        Expr(head, map(e->to_lowered_expr(mod, bindings, e), children(ex))...)
     end
 end
 
@@ -259,7 +259,7 @@ function Core.eval(mod::Module, ex::SyntaxTree)
         return x
     end
     linear_ir = lower(mod, ex)
-    expr_form = to_lowered_expr(mod, linear_ir.var_info, linear_ir)
+    expr_form = to_lowered_expr(mod, linear_ir.bindings, linear_ir)
     eval(mod, expr_form)
 end
 

src/kinds.jl

@@ -24,6 +24,8 @@ function _insert_kinds()
             "loopinfo"
             # Identifier for a value which is only assigned once
             "SSAValue"
+            # Unique identifying integer for bindings (of variables, constants, etc)
+            "BindingId"
             # Scope expressions `(hygienic_scope ex s)` mean `ex` should be
             # interpreted as being in scope `s`.
             "hygienic_scope"