3 Stars
Updated Last
3 Years Ago
Started In
January 2015

Rebuilding the Julia system image is time-consuming, leading to long round-trip times when modifying code in base. This tool aims to reduce the turn-around time for testing changes to inference.jl, the most compilation-heavy part of base.

This package should work (but has not been tested) with other modular parts of base, although it will almost certainly not work with the REPL code.


One can open a Julia REPL, navigate to base/, and then execute include("coreimg.jl"). This will replace the inference machinery in the running session. You can perform this multiple times as you insert and remove debugging statements in inference.jl.


using HackThatBase
func(x,y) = x + y
args = lminfo(func, (Int,Float64))   # (Int, Float64) are the types of x and y
@hack W inference
result = run_inference(W, args)

After modifying inference.jl, simply re-run these steps to execute the modified code:

@hack W inference
result = run_inference(W, args)

To view the resulting inferred AST, use

showast(args[1], result[2])

lminfo is a helper function to extract the method signature and other arguments to typeinf.

(on my system, the above steps take ~15 seconds to complete, as compared to >2 minutes to rebuild the second stage of sysimg)


  • imports are limited in the test environment. To inspect variables, push to an array in Main. e.g. somewhere in inference.jl, do push!(Main.foo, A) (where foo = [] at the REPL before running)

Tips on usage with Debug.jl

HackThatBase can be used in conjunction with the Debug package. Because inference is used during the compilation of functions (including those called by inference), it's advised that you first complete one run through showast before beginning debugging. For example, let's say you've edited inference.jl and inserted some breakpoints. Here's how you should proceed:

shell> git stash       # temporarily go back to the unedited version of inference.jl
julia> using HackThatBase
julia> func(x,y) = x + y
julia> args = lminfo(func, (Int,Float64))   # (Int, Float64) are the types of x and y
julia> @hack W inference
julia> result = run_inference(W, args)
julia> showast(args[1], result[1])
shell> git stash pop   # restore your edited version of inference.jl
julia> @hack W inference
julia> result = run_inference(W, args)

At this point your breakpoints will be triggered.