Find all variables referenced and defined in an expression. This package is used internally by Pluto to find links between cells.
julia> using ExpressionExplorer
julia> ex = :(const words = split(line));
julia> node = ExpressionExplorer.compute_reactive_node(ex);
julia> node.references
Set{Symbol} with 2 elements:
:line
:split
julia> node.definitions
Set{Symbol} with 1 element:
:words
This package (ExpressionExplorer.jl) is used by the package PlutoDependencyExplorer.jl, which is used by Pluto.jl.
ExpressionExplorer.jl looks at a single expression, and gives the definitions and references.
PlutoDependencyExplorer.jl looks at a series of expressions (i.e. all cells in a notebook) and gives the "topological order": the order in which to run cells, taking definitions and references into account.
The main function to use is compute_reactive_node(expression)
, which returns a ReactiveNode
. There is also a more low-level API available: compute_symbols_state
returning a SymbolsSate
.
If you are interested in the dependencies between expressions, then you should compute the ReactiveNode
for each expression. This is a data structure that looks like:
Base.@kwdef struct ReactiveNode
# core fields:
references::Set{Symbol} = Set{Symbol}()
definitions::Set{Symbol} = Set{Symbol}()
# more advanced fields:
soft_definitions::Set{Symbol} = Set{Symbol}()
funcdefs_with_signatures::Set{FunctionNameSignaturePair} = Set{FunctionNameSignaturePair}()
funcdefs_without_signatures::Set{Symbol} = Set{Symbol}()
macrocalls::Set{Symbol} = Set{Symbol}()
end
You can use the function compute_reactive_node(expression)
to explore an expression and generate the resulting ReactiveNode
.
Let's compute the ReactiveNode
for these two expressions:
julia> e1 = :(weather = magic() + science);
julia> e2 = :(weather() = magic() + science);
First one:
julia> r1 = ExpressionExplorer.compute_reactive_node(e1)
ExpressionExplorer.ReactiveNode(Set([:+, :magic, :science]), Set([:weather]), Set{Symbol}(), Set{ExpressionExplorer.FunctionNameSignaturePair}(), Set{Symbol}(), Set{Symbol}())
julia> r1.definitions
Set{Symbol} with 1 element:
:weather
julia> r1.references
Set{Symbol} with 3 elements:
:+
:magic
:science
julia> r1.funcdefs_without_signatures
Set{Symbol}()
Second one, note that weather
is a function definition, so it does not show up in r2.definitions
. If you want everything that is defined, you can use r2.definitions ∪ r2.funcdefs_without_signatures
.
julia> r2 = ExpressionExplorer.compute_reactive_node(e2)
ExpressionExplorer.ReactiveNode(Set([:+, :magic, :science]), Set{Symbol}(), Set{Symbol}(), Set(ExpressionExplorer.FunctionNameSignaturePair[ExpressionExplorer.FunctionNameSignaturePair([:weather], 0xa2e6e5b3d2eee6b5)]), Set([:weather]), Set{Symbol}())
julia> r2.definitions
Set{Symbol}()
julia> r2.references
Set{Symbol} with 3 elements:
:+
:magic
:science
julia> r2.funcdefs_without_signatures
Set{Symbol} with 1 element:
:weather
If you are not interested in just the dependencies between expressions, there is a more low-level data structure available. (We include it for completeness, but Pluto does not use this data, except to generate a ReactiveNode
.)
The function compute_symbols_state
take an expression as argument, and returns a SymbolsState
.
Base.@kwdef mutable struct SymbolsState
references::Set{Symbol} = Set{Symbol}()
assignments::Set{Symbol} = Set{Symbol}()
funccalls::Set{FunctionName} = Set{FunctionName}()
funcdefs::Dict{FunctionNameSignaturePair,SymbolsState} = Dict{FunctionNameSignaturePair,SymbolsState}()
macrocalls::Set{FunctionName} = Set{FunctionName}()
end
with
const FunctionName = Vector{Symbol}
struct FunctionNameSignaturePair
name::FunctionName
signature_hash::UInt
end
FunctionNameSignaturePair
looks like FunctionNameSignaturePair([:Base, :sqrt], UInt(0xb187232b478))
. It contains a "hash of the function signature, minus variable names", i.e. Base.sqrt(x::Int)::String
and Base.sqrt(x::Number)
will have different hashes, but Base.sqrt(x)
and Base.sqrt(woww)
won't.
julia> using ExpressionExplorer
julia> compute_symbols_state(:(a = b + c))
SymbolsState(
references=Set([:b, :c]),
assignments=Set([:a]),
funccalls=Set([[:+]]),
funcdefs=Dict{ExpressionExplorer.FunctionNameSignaturePair, SymbolsState}(),
macrocalls=Set{Vector{Symbol}}()
)
julia> compute_symbols_state(:(a = b))
SymbolsState(
references=Set([:b]),
assignments=Set([:a]),
funccalls=Set{Vector{Symbol}}(),
funcdefs=Dict{ExpressionExplorer.FunctionNameSignaturePair, SymbolsState}(),
macrocalls=Set{Vector{Symbol}}()
)
julia> compute_symbols_state(:(a(b) = b + c))
SymbolsState(
references=Set{Symbol}(),
assignments=Set{Symbol}(),
funccalls=Set{Vector{Symbol}}(),
funcdefs=Dict{ExpressionExplorer.FunctionNameSignaturePair, SymbolsState}(
ExpressionExplorer.FunctionNameSignaturePair([:a], 0x4e081629cf5e5d05) =>
SymbolsState(
references=Set([:c]),
assignments=Set{Symbol}(),
funccalls=Set([[:+]]),
funcdefs=Dict{ExpressionExplorer.FunctionNameSignaturePair, SymbolsState}(),
macrocalls=Set{Vector{Symbol}}()
)
),
macrocalls=Set{Vector{Symbol}}()
)
ExpressionExplorer ignores the arguments of macro calls. Macros can transform an expression into anything, so the output of ExpressionExplorer for expressions with a macro call is ambiguous. For example, the expression @time x
contains a reference to x
, while @gensym x
contains a definition of x
.
In this example, notice that the assignment to x
and reference to y
are detected, but AAA
and BBB
are ignored, because they happen inside a macro call argument.
julia> ExpressionExplorer.compute_reactive_node(quote
x = y
@time AAA = BBB
end)
ReactiveNode(Set([Symbol("@time"), :y]), Set([:x]), Set{Symbol}(), Set{FunctionNameSignaturePair}(), Set{Symbol}(), Set([Symbol("@time")]))
You can check whether there were any unexplored macro call arguments with the .macrocalls
field of the ReactiveNode
, which should be an empty set.
To solve this, you can macroexpand expressions before giving them to ExpressionExplorer. For example:
julia> ExpressionExplorer.compute_reactive_node(macroexpand(Main, quote
x = y
@time AAA = BBB
end))
ReactiveNode(Set([:first, :GC_Diff, :isnothing, :gc_alloc_count, :-, :gc_num, :cumulative_compile_time_ns, :time_ns, :y, :BBB, :print, :cumulative_compile_timing, :time_print, :last, :!]), Set([:AAA, :x]), Set{Symbol}(), Set{FunctionNameSignaturePair}(), Set{Symbol}(), Set{Symbol}())
Notice that now, AAA
and BBB
are detected, along with functions used inside the @time
expression.
You can also ignore all macros, and analyse the macro call arguments as if the macro was not there. Here is a discussion showing how to do it.
The package also includes some utility functions used by Pluto.jl, that might also be useful to other packages.
With compute_usings_imports
you can extract all using
or import
expressions contained in a larger expression.
julia> ex = quote
if something
import A.B: c
else
using D
end
end
quote;
julia> result = compute_usings_imports(ex);
julia> result.usings
Set{Expr} with 1 element:
:(using D)
julia> result.imports
Set{Expr} with 1 element:
:(import A.B: c)
This function is used by Pluto's built-in package manager to learn which packages are used in a notebook.
get_rootassignee(ex)::Union{Symbol,Nothing}
If the expression is a (simple) assignment at its root, return the assignee as Symbol
, return nothing
otherwise.
is_toplevel_expr(ex)::Bool
Return whether the expression is of the form Expr(:toplevel, LineNumberNode(..), any)
.