Skip to content

Latest commit

 

History

History
69 lines (47 loc) · 1.64 KB

2021-notes.md

File metadata and controls

69 lines (47 loc) · 1.64 KB
Raw

README from 2021

  • We know that Julia subtyping is undecidable.
  • Can we restrict lower bounds to make subtyping decidable, at the same time allowing all the practical uses of lower bounds in the existing Julia code?

Using static analysis, we find uses of lower bounds in Julia packages and manually inspect them.

Static analysis of lower bounds

As of Julia 1.5.3, there are only two accepted patterns of lower bounds in the full form of where-types,

  1. where T >: Int
  2. where Int <: T <: Number

as well as one shorthand form Vector{>:Int} (which means Vector{T} where T>:Int).

Everything else (e.g. where Int <: T) doesn't work, and test/lb-analysis.jl tests for that in the tests set "lb-analysis :: type bounds format".

The three patterns can be found in extractLowerBound function, src/lb-analysis/process-code.jl.

Note. Not all lower bounds that we find are in function definitions because we don't specifically match where, just T >: LB. Thus, we also find run-time checks for lower bounds.

Getting the data

Assuming the directory structure:

.
|
-- julia-sub
-- utils
     |
     -- JuliaPkgsList.jl
     -- JuliaPkgDownloader

Old stuff below

and jl-wa with a clonning script

$ ../../utils/JuliaPkgsList.jl/gen-pkgs-list.jl 0 -o pkgs-list/pkgs-list.txt -r

$ julia -p 8 ../../jl-wa/src/utils/clone.jl -s pkgs-list/pkgs-list.txt -d pkgs/4886/

Running the analysis

$ julia analysis-script.jl <pkgs>

where <pkgs is a folder with Julia packages.

lia analysis-script.jl 


where `<pkgs` is a folder with Julia packages.