Awesome
SoftGlobalScope
SoftGlobalScope is a package for the Julia language that simplifies the variable scoping rules for code in global scope. It is intended for interactive shells (the REPL, IJulia, etcetera) to make it easier to work interactively with Julia, especially for beginners.
In particular, SoftGlobalScope provides a function softscope
that can transform Julia code from using the default "hard" scoping rules to simpler "soft" scoping rules in global scope only.
In Julia 1.5, "soft scoping" became the default for the built-in Julia REPL, and the SoftGlobalScope module simply calls through to the standard-library REPL.softscope
function. The advantage of using the SoftGlobalScope library is that it also works with previous Julia versions.
Hard and soft global scopes
Starting in Julia 0.7, when you assign to global variables in the context of an inner scope (a for
loop or a let
statement) you need to explicitly declare the variable
as global
in order to distinguish it from declaring a new variable. We refer to this as "hard" scoping rules. For example, the following code gives an warning in 0.7:
julia> s = 0
0
julia> for i = 1:10
s = s + i
end
┌ Warning: Deprecated syntax `implicit assignment to global variable `s``.
└ Use `global s` instead.
and an error in Julia 1.0:
julia> s = 0
0
julia> for i = 1:10
s = s + i # wrong: defines a new local variable s
end
ERROR: UndefVarError: s not defined
To make it work in 1.0, you need a global
declaration:
julia> for i = 1:10
global s = s + i
end
julia> s # should be 1 + 2 + ⋯ + 10 = 55
55
This only applies to global variables; similar code inside a function (or whenever s
is a local variable) works fine without any added keyword:
julia> function f(n)
s = 0
for i = 1:n
s = s + i
end
return s
end
f (generic function with 1 method)
julia> f(10)
55
There were various reasons for this scoping rule, e.g. to facilitate static analysis by the compiler, and it isn't too onerous in "serious" Julia code where little code executes in global scope.
However, for interactive use, especially for new users, the necessity of the global
keyword, and the difference between code in local and global scopes, can be confusing and was ultimately reverted. The SoftGlobalScope package exists to make it easier for interactive shells to automatically insert the global
keyword in common cases, what we term "soft" global scope.
Usage
The SoftGlobalScope
module exports two functions softscope
and softscope_include_string
, and a macro @softscope
:
You can transform the expression using softscope(module, expression)
to automatically insert the necessary global
keyword. For example, assuming that the module Main
has a global variable s
(as above), you can do:
julia> softscope(Main, :(for i = 1:10
s += i
end))
:(for i = 1:10
#= REPL[3]:2 =#
global s += i
end)
You can then execute the statement with eval
. Alternatively, you can decorate the expression with the @softscope
macro:
julia> s = 0;
julia> @softscope for i = 1:10
s += i
end
julia> s
55
This macro should only be used in the global scope (e.g., via the REPL); using this macro within a function is likely to lead to unintended consequences.
You can execute an entire sequence of statements using "soft" global scoping rules via softscope_include_string(module, string, filename="string")
:
julia> softscope_include_string(Main, """
s = 0
for i = 1:10
s += i
end
s
""")
55
(This function works like include_string
, returning the value of the last evaluated expression.)
In Julia 0.6, no code transformations are required, so softscope
returns the original expression
and softscope_include_string
is equivalent to include_string
. In Julia 1.5, softscope_include_string(m, x)
is equivalent to include_string(REPL.softscope, m, x)
using the softscope
transformation function provided by Julia's REPL
standard library.
Contact
SoftGlobalScope was written by Steven G. Johnson and is free/open-source software under the MIT/Expat license. Please file bug reports and feature requests at the SoftGlobalScope github page.