Assume effects

[Tokazama]

Fixes #769

This allows packages with a 1.6 compat to replace Base.@pure with Base.@assume_effects :total similar to what's been done in base. All other effects are ignored

[Keno]

@pure is stronger than @assume_effects. I don't think this is a good idea.

[codecov]

Codecov Report

Merging #783 (b525933) into master (6e1b40c) will increase coverage by 0.06%.
The diff coverage is 100.00%.

@@            Coverage Diff             @@
##           master     #783      +/-   ##
==========================================
+ Coverage   91.63%   91.69%   +0.06%     
==========================================
  Files           2        2              
  Lines         251      253       +2     
==========================================
+ Hits          230      232       +2     
  Misses         21       21              

Impacted Files	Coverage Δ
src/Compat.jl	92.24% <100.00%> (+0.06%)	⬆️

Help us with your feedback. Take ten seconds to tell us how you rate us. Have a feature suggestion? Share it here.

[Tokazama]

@pure is stronger than @assume_effects. I don't think this is a good idea.

Sorry, I thought this was how all uses of Base.@pure were managed in Base. What's the appropriate way to transition code that uses Base.@pure?

[Keno]

It should be generally safe to replace calls to @pure, by @assume_effects :total, but not the other way around. The issue is that with this, you'll write @assume_effects :total, but get @pure semantics on older versions of julia, which is not the same.

[Keno]

More generally, though, packages should revisit every use of @pure and select the minimum set of effect assumptions that they need (preferably none).

[Tokazama]

I removed the mapping to @pure so there aren't any added assumptions on 1.6.
I guess there could be an extended help section for Base.@pure detailing how to make the transition but a quick search on JuliaHub reveals that most packages that still use it are maintained by one particular individual. If I recall correctly, previous attempts to dissuade their use of @pure were met with quite a bit of resistance.

[martinholters]

Needs a README entry.

[martinholters]

LGTM, but we should get #784 in first and then test on top of that to verify it's working as intended on latest Julia.

That said, I wonder how helpful this is given it cannot be used as a replacement for @pure without regressions on older Julia. Still better than nothing, though.

@Keno any more thoughts on this?

[martinholters]

Can you rebase on (or merge) current master to (hopefully) fix the test failures on nightly?