@code_expr fails (?) with a method from IntervalArithmetic.jl

[dpsanders]

julia> using CodeTracking, Revise

julia> using IntervalArithmetic

julia> Revise.track(IntervalArithmetic)

julia> @code_expr +(0.1, 0.2, RoundDown)
┌ Warning: Tuple{typeof(+), T, T, RoundingMode} where T<:AbstractFloat was not found
└ @ Revise ~/.julia/packages/Revise/9lZUE/src/packagedef.jl:1083

[timholy]

Sadly, Revise can't handle @eval statements hidden within function bodies, only those at top-level. I think that accounts for fewer than 20 of the >18k methods in Base, so it's never been a priority.

[timholy]

BTW you don't need that track statement; all your using and import packages get tracked automatically.

[dpsanders]

OK, I see, thanks. Unfortunately that's pretty key to the functioning of the package as it's currently written.

[KristofferC]

🏴‍☠️

[timholy]

I don't think there's even a safe & comprehensive fix for this without moving Revise's core method-tracking functionality into julia proper; otherwise Revise would have to go around executing random functions with random arguments just to see if they happen to define methods.

Interestingly, I actually think the time for such a move may have come; there just seem to be an endless list of things that would benefit from having cached "fenced" source code and expressions a la JuliaLang/julia#31162. It would fix most of Julia's usability gaps (stacktraces, debugger, integration of type inference results into IDEs) as well as the main source of Revise bugs (lowering always generates new gensyms). So I'm not saying it will never be supported, but it's a monumental change.

I'll close this as a won't fix because this basically can't be tackled as an independent step.

[timholy]

I don't think there's even a safe & comprehensive fix for this without moving Revise's core method-tracking functionality into julia proper; otherwise Revise would have to go around executing random functions with random arguments just to see if they happen to define methods.

Interestingly, I actually think the time for such a move may have come; there just seem to be an endless list of things that would benefit from having cached "fenced" source code and expressions a la JuliaLang/julia#31162. It would fix most of Julia's usability gaps (stacktraces, debugger, integration of type inference results into IDEs) as well as the main source of Revise bugs (lowering always generates new gensyms). So I'm not saying it will never be supported, but it's a monumental change.

I'll close this as a won't fix because this basically can't be tackled as an independent step.