cako / PackageCompiler.jl

Compile your Julia Package

PackageCompiler

Remove just-in-time compilation overhead from your package and compile it into a system image.

Usage example

E.g. do:

using PackageCompiler

# This command will use the `runtest.jl` of `ArgParse` + `SnoopCompile` to find out what functions to precompile!
# `force = false` to not force overwriting Julia's current system image
compile_package("ArgParse", "SnoopCompile", force = false, reuse = false)

# Build again, reusing the snoop file
compile_package("ArgParse", "SnoopCompile", force = false, reuse = true)

# You can define a file that will get run for snooping explicitly like this:
# this makes sure, that binary gets cached for all functions called in `for_snooping.jl`
compile_package(("ArgParse", "relative/path/for_snooping.jl"))

# If you used force and want your old system image back (force will overwrite the default system image Julia uses) you can run:
revert()

# Or if you simply want to get a native system image e.g. when you have downloaded the generic Julia install:
force_native_image!()

# Build an executable
build_executable(
    "hello.jl", # Julia script containing a `julia_main` function, e.g. like `examples/hello.jl`
    snoopfile = "call_functions.jl", # Julia script which calls functions that you want to make sure to have precompiled [optional]
    builddir = "path/to/builddir" # that's where the compiled artifacts will end up [optional]
)

# Build a shared library
build_shared_lib("hello.jl")

Troubleshooting:

• You might need to tweak your runtest, since SnoopCompile can have problems with some statements. Please open issues about concrete problems! This is also why there is a way to point to a file different from runtests.jl, for the case it becomes impossible to combine testing and snoop compiling, just pass ("package", "snoopfile.jl")!

• Non constant globals and globals defined in functions are problematic. Removing those got me to 95% of making the package safe for static compilation.

• Type unstable code had some inference issues (around 2 occurrence, where I’m still not sure what was happening, and both cases happened with dictionaries). The only way to find those was investigating the segfaults with gdb, but then it was relatively easy to just juggle around the code, since the stacktraces accurately pointed to the problem. The non constant globals might be related since they introduce type instabilities.

• Some generated functions needed reordering of the functions they call (actually, even for normal compilation, all functions that get called in a generated function should be defined before it).

• I uncovered one out-of-bounds issue, that somehow was not coming up without static compilation.

• I used julia-debug to uncover most bugs, but actually the last errors I was trying to uncover where due to julia-debug itself!

• You’re pretty much on your own and need to use gdb to find any issues and I still don’t know what the underlying julia issues are and when they will get fixed 😉 See: JuliaLang/julia#24533. Hopefully we'll look at a better story with Julia 1.0!

Static Julia Compiler

Build shared libraries and executables from Julia code.

Run juliac.jl -h for help:

usage: juliac.jl [-v] [-q] [-d <dir>] [-n <name>] [-p <file>] [-c]
                 [-a] [-o] [-s] [-i] [-e] [-t] [-j] [-f <file>] [-r]
                 [-R] [-J <file>] [-H <dir>] [--startup-file {yes|no}]
                 [--handle-signals {yes|no}]
                 [--sysimage-native-code {yes|no}]
                 [--compiled-modules {yes|no}]
                 [--depwarn {yes|no|error}]
                 [--warn-overwrite {yes|no}]
                 [--compile {yes|no|all|min}] [-C <target>]
                 [-O {0,1,2,3}] [-g <level>] [--inline {yes|no}]
                 [--check-bounds {yes|no}] [--math-mode {ieee,fast}]
                 [--cc <cc>] [--cc-flags <flags>] [--version] [-h]
                 juliaprog [cprog]

Static Julia Compiler

positional arguments:
  juliaprog             Julia program to compile
  cprog                 C program to compile (required only when
                        building an executable, if not provided a
                        minimal driver program is used)

optional arguments:
  -v, --verbose         increase verbosity
  -q, --quiet           suppress non-error messages
  -d, --builddir <dir>  build directory
  -n, --outname <name>  output files basename
  -p, --snoopfile <file>
                        specify script calling functions to precompile
  -c, --clean           remove build directory
  -a, --autodeps        automatically build required dependencies
  -o, --object          build object file
  -s, --shared          build shared library
  -i, --init-shared     add `init_jl_runtime` and `exit_jl_runtime` to
                        shared library for runtime initialization
  -e, --executable      build executable file
  -t, --rmtemp          remove temporary build files
  -j, --copy-julialibs  copy Julia libraries to build directory
  -f, --copy-file <file>
                        copy file to build directory, can be repeated
                        for multiple files
  -r, --release         build in release mode, implies `-O3 -g0`
                        unless otherwise specified
  -R, --Release         perform a fully automated release build,
                        equivalent to `-atjr`
  -J, --sysimage <file>
                        start up with the given system image file
  -H, --home <dir>      set location of `julia` executable
  --startup-file {yes|no}
                        load `~/.julia/config/startup.jl`
  --handle-signals {yes|no}
                        enable or disable Julia's default signal
                        handlers
  --sysimage-native-code {yes|no}
                        use native code from system image if available
  --compiled-modules {yes|no}
                        enable or disable incremental precompilation
                        of modules
  --depwarn {yes|no|error}
                        enable or disable syntax and method
                        deprecation warnings
  --warn-overwrite {yes|no}
                        enable or disable method overwrite warnings
  --compile {yes|no|all|min}
                        enable or disable JIT compiler, or request
                        exhaustive compilation
  -C, --cpu-target <target>
                        limit usage of CPU features up to <target>
                        (implies default `--sysimage-native-code=no`)
  -O, --optimize {0,1,2,3}
                        set the optimization level (type: Int64)
  -g, --debug <level>   enable / set the level of debug info
                        generation (type: Int64)
  --inline {yes|no}     control whether inlining is permitted
  --check-bounds {yes|no}
                        emit bounds checks always or never
  --math-mode {ieee,fast}
                        disallow or enable unsafe floating point
                        optimizations
  --cc <cc>             system C compiler
  --cc-flags <flags>    pass custom flags to the system C compiler
                        when building a shared library or executable
  --version             show version information and exit
  -h, --help            show this help message and exit

examples:
  juliac.jl -vae hello.jl        # verbose, build executable and deps
  juliac.jl -vae hello.jl prog.c # embed into user defined C program
  juliac.jl -qo hello.jl         # quiet, build object file only
  juliac.jl -vosej hello.jl      # build all and copy Julia libs
  juliac.jl -vRe hello.jl        # fully automated release build

Building a shared library

PackageCompiler can compile a julia library into a linkable shared library, built for a specific architecture, with a C-compatible ABI which can be linked against from another program. This can be done either from the julia api, build_shared_lib("src/HelloLib.jl", "hello"), or on the command line, $ juliac.jl -vas src/HelloLib.jl. This will generate a shared library called builddir/libhello.{so,dylib,dll} depending on your system.

The provided julia file, src/HelloLib.jl, is PackageCompiler's entry point into the library, so it should be the "top level" library file. Any julia code that it includes or imports will be compiled into the shared library.

Note that for a julia function to be callable from C, it must be defined with Base.@ccallable, e.g. Base.@ccallable foo()::Cint = 3.

Building an executable

To compile a Julia program into an executable, you can use either the julia api, build_executable("hello.jl", "hello"), or the command line, $ juliac.jl -vae hello.jl.

The provided julia file, hello.jl, is PackageCompiler's entry point into the program, and should be the program's "main" file. Any julia code that it includes or imports will be compiled into the shared library, which will be linked against the provided C program to create an executable at builddir/hello.

If you choose to use the default C program, your julia code must define julia_main as its entry point. The resultant executable will start by calling that function, so all of your program's logic should proceed from that function. For example:

Base.@ccallable function julia_main(ARGS::Vector{String})::Cint
    hello_main(ARGS)  # call your program's logic.
    return 0
end

Please see examples/hello.jl for an example Julia program.

Notes

1. The juliac.jl script is located in the PackageCompiler root folder (normpath(Base.find_package("PackageCompiler"), "..", "..")).

2. A shared library containing the system image hello.so, and a driver binary hello are created in the builddir directory. Running hello produces the following output:

   hello, world
   sin(0.0) = 0.0
      ┌─────────────────────────────────────────────────┐
    1 │⠀⠀⠀⠀⠀⠀⠀⡠⠊⠉⠉⠉⠢⡀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀│
      │⠀⠀⠀⠀⠀⢠⠎⠀⠀⠀⠀⠀⠀⠘⢆⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀│
      │⠀⠀⠀⠀⢠⠃⠀⠀⠀⠀⠀⠀⠀⠀⠀⠳⡀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀│
      │⠀⠀⠀⢠⠃⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠱⡀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀│
      │⠀⠀⢠⠃⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠳⡀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀│
      │⠀⢀⠇⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⢣⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀│
      │⠀⡎⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⢇⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀│
      │⠼⠤⠤⠤⠤⠤⠤⠤⠤⠤⠤⠤⠤⠤⠤⠤⠤⠤⠤⠬⢦⠤⠤⠤⠤⠤⠤⠤⠤⠤⠤⠤⠤⠤⠤⠤⠤⠤⠤⢤│
      │⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠈⡆⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⢀⠇│
      │⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠘⡄⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⡎⠀│
      │⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠱⡀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⡞⠀⠀│
      │⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠱⡀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⡜⠀⠀⠀│
      │⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠱⡀⠀⠀⠀⠀⠀⠀⠀⠀⠀⡞⠀⠀⠀⠀│
      │⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠘⢆⠀⠀⠀⠀⠀⠀⢠⠎⠀⠀⠀⠀⠀│
   -1 │⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠑⢄⣀⣀⣀⠔⠁⠀⠀⠀⠀⠀⠀│
      └─────────────────────────────────────────────────┘
      0                                               100

3. Currently, before another program can call any of the functions defined in the created shared library, that program must first initialize the julia runtime. (See #53 for details.)

Under the hood

The juliac.jl script uses the --output-o switch to compile the user script into object code, and then builds it into the system image specified by the -J switch. This prepares an object file, which is then linked into a shared library containing the system image and user code. A driver script such as the one in program.c can then be used to build a binary that runs the Julia code.

Instead of a driver script, the generated system image can be embedded into a larger program, see the Embedding Julia section of the Julia manual. Note that the name of the generated system image ("libhello" for hello.jl) is accessible from C in the preprocessor macro JULIAC_PROGRAM_LIBNAME.

For more information on static Julia compilation see:
https://juliacomputing.com/blog/2016/02/09/static-julia.html