fork4jl / buildsystem

BuildSystem : A simple experimental makefile system for Julia

Instructions on how to use BuildSystem package

To use this package, you must use our modified version of Julia compiler.

Due to some technical limitations, Julia fails to build system image in our modified compiler. So we need to first build the system image in an unmodified Julia, then apply our patch.

Build a modified Julia

Firstly download our fork of Julia https://github.com/ChenNingCong/julia.git, then checkout to commit 55808a56e0fcd147b10e35d64721a1849c841196. Build Julia normally following official documents. Then switch to master and run make julia-noimage-release. This will build julia without running system image build process. After this you can run make binary-dist to get a packed julia distribution (not necessary).

Build a binary cacche

Firstly ensure that you install this package BuildSytem in the same directory where you install you julia executable. For example, if you unpack your julia installation (or build) at path /home/username/julia, then you also need to install the package with path /home/username/BuildSystem.

To run the demo, follow these steps:

1. At root path of BuildSystem, run source test/help.sh in shell. This script creates two shortcuts ju18 and jbuild. ju18 is used to start julia in the image-codegen mode, and jbuild drives the makefile system for julia.
2. At root path of BuildSystem, run jbuild test/json_ninja.jl. This will build a compiled cache of JSON package. You need to wait a couple of time because we need to firstly compile BuildSystem, Test and Pkg, on which the build system depends. You need to compile them only once because the results are cached and can be reused. The actual time spend on compiling JSON is small.
3. If you encounter no error at step 1 and 2, you can check the directory of test/binary, where you can find a bundle of compiled binary files.

Now you have the compiled binary generated from test files, it's time to use these binaries!

1. At root path of BuildSystem, run ju18. This command will start a new Julia (version 1.8.0-DEV).
2. Pkg.activate("test") to activate the environment at test folder. Run Pkg.instantiate() to install dependencies. This may take some time, for Pkg.instantiate is not compiled due to some technical limitation.
3. Now excecute @use JSON, to import both the package and the binary.
4. Run x = JSON.parse("123"), you can notice that the latency is completely gone.
5. Run

   s = "{\"a_number\" : 5.0, \"an_array\" : [\"string\", 9]}"
   JSON.parse(s)
   

6. You may notice that step 5 still have latency. This is largely due to the compilation of display in the REPL, which is generally not touched in test file. To conquer this problem, run saveWork([:JSON]). This will magically save all the compiled code in JSON generated in this REPL session.
7. To use the REPL cache, the next time you open REPL, after you load JSON library (after step 3, this is very important), run loadREPL() to bring in the binary cache you save, repeat step 5,6 and now you can see that the latency is completely gone!!!