Skip to content

LilithHafner/PtrArrays.jl

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

23 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

PtrArrays

Build Status Coverage PkgEval Aqua deps

Do you miss playing hide and seek with memory leaks? Do you find GC overhead problematic? PtrArrays.jl can take you back to the good old days of manual memory management. See also Bumper.jl if you want to avoid GC overhead and don't like hide and seek.

This package provides malloc(T, dims...) which allocates an AbstractArray{T} with the provided dims. If you want, you can call free on the array once you're done using it but it can be more fun to see what happens if you don't.

Example usage

julia> malloc(Int, 4)
4-element PtrArray{Int64, 1}:
 1053122630
          0
  936098496
  936099008

julia> free(ans)

julia> malloc(Int, 4, 4)
4×4 PtrArray{Int64, 2}:
       923300075       1046634192       1046634192       1046634408
               0              120              124              152
               0                0                0                0
 281474587621896  281474587621899  281474587621900  281474587621896

julia> free(ans)

Benchmarks:

using PtrArrays
function f(n)
    x = malloc(Int, n)
    try
        sum(x) # Let's see what we get!
    finally
        free(x) # Putting the `free` call in a finally block makes memory leaks less common
    end
end

f(1000)
# 6474266410623015

using Chairmarks
@b f(1000)
# 101.317 ns

function g(n)
    x = Vector{Int}(undef, n)
    sum(x) # Let's see what we get!
end

@b g(1000)
# 130.125 ns (3 allocs: 7.875 KiB)

The whole package's source code is only about 44 lines (excluding comments and whitespace), half of which is re-implementing Julia's buggy Core.checked_dims function. Read it here

Alternatives

Bumper.jl provides bump allocators which allow you to manage your own allocation stack, bypassing the Julia GC.

StaticTools.jl is a much larger package which provides, among other things, a MallocArray type that behaves similarly to PtrArray. As StaticTools.jl is meant to run without the Julia runtime, it does not make use of features such as exceptions. Consequently, all usages of StaticTools.MallocArray are implicitly annotated with @inbounds and out of bounds accesses are UB instead of errors, StaticTools.MallocArray with invalid indices will return a null pointer with size zero while PtrArrays will throw, etc. In general, StaticTools.jl's MallocArray can be thought of as "unsafe" while PtrArray is "safe" (though memory leaks will still occur if you call malloc and fail to call free)

MallocArrays.jl is the original name of this package. It is obsolete.