Skip to content

RelationalAI-oss/Blobs.jl

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Blobs

Blobs makes it easy to lay out complex data-structures within a single memory region. Data-structures built using this library:

  • are relocatable - internal pointers are converted to offsets, so the entire memory region can be written to / read from disk or sent over the network without pointer patching
  • require no deserialization - they can be directly read/written without first copying the data into a Julia-native data-structure
  • require no additional heap allocation - field access is just pointer arithmetic and every field read/write returns an isbitstype type which can stored on the stack

This makes them ideal for implementing out-of-core data-structures or for DMA to co-processors.

Safety

This library does not protect against:

  • giving an incorrect length when creating a Blob
  • using a Blob after freeing the underlying allocation

Apart from that, all other operations are safe. User error or invalid data can cause AssertionError or BoundsError but cannot segfault the program or modify memory outside the blob.

Usage

Acquire a Ptr{Nothing} from somewhere:

julia> struct Foo
       x::Int64
       y::Bool
       end

julia> p = Libc.malloc(sizeof(Foo))
Ptr{Nothing} @0x0000000006416020

We can interpret this pointer as any isbitstype Julia struct:

julia> foo = Blob{Foo}(p, 0, sizeof(Foo))
Blobs.Blob{Foo}(Ptr{Nothing} @0x0000000004de87c0, 0, 16)

(See Blobs.malloc_and_init for a safer way to create a fresh blob).

We can access references to fields of Foo using the fieldnames directly:

julia> foo.x
Blobs.Blob{Int64}(Ptr{Nothing} @0x0000000004de87c0, 0, 16)

julia> foo.y
Blobs.Blob{Bool}(Ptr{Nothing} @0x0000000004de87c0, 8, 16)

And use [] to derefence Blobs:

julia> foo[]
Foo(114974496, true)

julia> foo.x[]
114974496

julia> foo.y[]
true

julia> y = foo.y
Blobs.Blob{Bool}(Ptr{Nothing} @0x0000000004de87c0, 8, 16)

julia> y[]
true

Similarly for setting values:

julia> foo[] = Foo(12, true)
Foo(12, true)

julia> foo[]
Foo(12, true)

julia> foo.y[] = false
false

julia> foo.y[]
false

julia> x = foo.x
Blobs.Blob{Int64}(Ptr{Nothing} @0x0000000004de87c0, 0, 16)

julia> x[] = 42
42

julia> x[]
42

julia> foo.x[]
42

The various data-structures provided can be nested arbitrarily. See the tests for examples.

Compatibility with Previous Versions

In the previous versions of this library, two macros @v and @ were used and we keep them for compatibility reasons. This macros bypass some of the bound-checkings and safety measures that are in-place in the normal usage of Blobs. In this section, we will introduce their usage.

Assume that we have the following Foo struct:

julia> struct Foo
       x::Int64
       y::Bool
       end

julia> m = Blobs.malloc_and_init(Foo)
Blob{Foo}(Ptr{Nothing} @0x00007fa0b84234e0, 0, 9)

Use the @a (for address) macro to obtain pointers to the fields of this struct:

julia> @a m.x
Blob{Int64}(Ptr{Nothing} @0x00007fa0b84234e0, 0, 9)

julia> @a m.y
Blob{Bool}(Ptr{Nothing} @0x00007fa0b84234e0, 8, 9)

Or the @v (for value) macro to dereference those pointers:

julia> @v m.x
44307392

julia> @v m.y
false

julia> y = @a m.y
Blob{Bool}(Ptr{Nothing} @0x00007fa0b84234e0, 8, 9)

julia> @v y
false

The @v macro also allows setting the value of a pointer:

julia> @v m.y = true
true

julia> @v m.y
true

julia> x = @a m.x
Blob{Int64}(Ptr{Nothing} @0x00007fa0b84234e0, 0, 9)

julia> @v x = 42
42

julia> @v x
42

julia> @v m.x
42

Also in this case, data-structures can be nested arbitrarily. See the compat-tests for examples.