Modifyfield.jl

Creates functions to modify immutable fields of a composite type inside a container
Author StephenVavasis
Sub Category Composite Data Types
Category File IO
Github
Popularity
3 Stars
Updated Last
7 Years Ago
Started In
May 2015

Modifyfield package

This Julia package provides macros @modify_field!, @modify_fields! and @modify_tuple_entry!. Macro @modify_field! is intended to modify a field of an object of an immutable composite type that sits inside of a container. To illustrate the issue, consider the following immutable structure:

immutable Immut
    intfld::Int
    isadded::Bool
end

Suppose a is an array of type Immut and the following loop is needed:

# LOOP 1
t = 0
for k = 1 : n
    t += a[k].intfld
    a[k].isadded = true
end

Unfortunately, the above code snippet is illegal because it modifies a field of an immutable object. [Aside: If Immut had been declared as a plain composite type with a type declaration rather than immutable, then the above code LOOP 1 would be fine. However, Julia programmers often put immutable composite types into containers rather than plain composite types because the former are packed densely in memory which may yield better performance.] Instead, we could obtain the same effect legally via:

# LOOP 2
t = 0
for k = 1 : n
    t += a[k].intfld
    a[k] = Immut(a[k].intfld, true)
end

The problem with this code is that it becomes unwieldy for a composite type with many fields. In this case, it would be hard to read and also a possible source of bugs if the arguments to the Immut constructor were ordered incorrectly.

To use the macro in the above example, first include the declaration using Modifyfield and then write:

# LOOP 3
t = 0
for k = 1 : n
    t += a[k].intfld
    @modify_field! a[k].isadded = true
end

Thus, the @modify_field! macro allows for code that mimics the clean syntax of LOOP 1 above while "under the hood" providing an implementation equivalent to LOOP 2 above.

The original version of this code was by S. Vavasis and used metaprogramming and the Val and Type types of Julia for dispatching to the correct routine. It was greatly improved by Simon Byrne with the incorporation of macros and generated functions.

If the user prefers to invoke a function rather than a macro, he/she can use the following statement for the same effect:

# LOOP 4
t = 0
for k = 1 : n
    t += a[k].intfld
    a[k] = copy_and_modify(a[k], Val{:isadded}, true)
end

Note that although the macro is intended for immutable objects in a container, it also works for immutable objects bound to a plain Julia variable:

julia> using Modifyfield.@modify_field!

julia> y = Immut(6,false)
Immut(6,false)

julia> @modify_field! y.intfld = 9
Immut(9,false)

However, for composite types that do not occur inside of larger containers, higher performance is usually attained by declaring objects such as this as a type rather than immutable especially if one is frequently modifying fields. (A type rather than immutable is also stylistically preferred in this context.)

Modifying multiple fields

A macro is also provided for modifying multiple fields at the same time. (This is more efficient than modifying one at a time.) Here is an example of its usage:

immutable Immut2
    intfld::Int
    isadded::Bool
    xx::Float64
end

If a is an array of Immut2 entries, then the following loop changes the first two fields of each entry:

for k = 1 : n
    @modify_fields! a[k].(intfld = k+1, isadded = true)
end

This slightly odd syntax was chosen so that field names are close to their corresponding new values to improve readability.

The parenthesized argument in the @modify_fields! macro can name a single field, but in this case it should be followed by a comma (so that its syntax matches the Julia tuple syntax):

@modify_fields! w.(intfld = 6,)

which is equivalent to:

@modify_field! w.intfld = 6

Modifying tuple entries

Similarly, the package provides a macro for modifying tuple entries. Here is an example of its execution:

julia> using Modifyfield.@modify_tuple_entry!

julia> t = (5,9.5,true)
(5,9.5,true)

julia> @modify_tuple_entry! t[2] = false
(5,false,true)

There is also an equivalent functional call:

julia> using Modifyfield.copy_and_modify_tup

julia> t = (5,9.5,true)
(5,9.5,true)

julia> t = copy_and_modify_tup(t, Val{2}, true)
(5,true,true)

As in the case of immutables, the implementation of @modify_tuple_entry! actually copies the entire tuple over.

A couple of cautionary notes are in order. First, the macro @modify_tuple_entry! requires a literal integer for the subscript (which is 2 in the above example) of the tuple. A variable or more general expression may not be used. The function-call version copy_and_modify_tup can take a variable subscript, e.g., Val{j} as its second argument, but this leads to a loss of performance because the compiler cannot fully determine argument types, and therefore the method dispatch happens at run time.

Second, the main purpose of this macro is for tuples that are packed inside of some other container in a high-performance setting. If one is modifying bare tuples such as t in the above example, then in most cases a cell array (Array{Any,1}) would be preferable to a tuple.