dev harness for inference?

[Isaiah Norton]

Is there any way to work on inference.jl without recompiling the whole sysimage after every change? I don't want the changes to actually be used by the compiler, just to make a change and run typeinf to see the result (or debugging output).

Not sure if this is feasible yet, but I was hoping to be able to `include("inference.jl")` in a baremodule with a sufficient number of imports from Base and Core.Intrinsics. Importing all of `names(Base,true)` won't work because the existing methods are overwritten and everything breaks. I have a few ideas for how to filter the existing inference methods out, but any suggestions on this (or anything else to the same end!) would be appreciated.

[Isaiah Norton]

Alright, I have something which seems to work in the simple case:

https://github.com/ihnorton/HackThatBase.jl

It works by parsing the given file and extracting all top-level declarations, and using that as an exclusion list for Base imports.

Example: https://gist.github.com/ihnorton/4c3b345adbdf7c549d95

In the above example I added `@show linfo` at the start of the main typeinf method.

The @hack loading step takes about 2 seconds, so ~15s total. By comparison, `make -j2` takes about 3 minutes on my computer after making any changes to inference.jl (and I'm pretty sure @show and similar calls aren't available).