Tag MacroTools.jl v0.3.5

[attobot]

Repository: MikeInnes/MacroTools.jl
Release: v0.3.5
Travis:
Diff: vs v0.3.4
requires vs v0.3.4: no changes
cc: @MikeInnes

[tkelman]

This has started causing segfaults in packages that depend on DataFrames on Julia 0.5.

[abelsiqueira]

Also segfaults for packages depending on Lazy.jl

[timholy]

It doesn't appear to cause the problem on release-0.5 (0.5.1-pre+31). Pretty please, can we have a julia 0.5.1?

[tkelman]

I'm working on it, but yes, it does. It's intermittent.

[tkelman]

My mistake, I was reproducing via Mimi.jl which was actually shelling out to whatever julia happened to be on the path, which wasn't the one I thought I was testing against. Even if I get 0.5.1 released soon, 0.5.0 is still widely used and this release should be reverted for now. Either marked 0.6-only, or replaced with an 0.3.6 that has the same sha as 0.3.4.

[aviks]

PR #7921 reverts this.

[tkelman]

it shouldn't just be reverted, but superseded

[tkelman]

or have its requirements adjusted (which would possibly require a test-list exemption, depending which direction we do it in)

[tkelman]

We have 3 immediate options here, the faster we do one of them the better.

1. tag a MacroTools 0.3.6 using the same commit as 0.3.4 - this needs someone with commit access to create the git tag to pass travis, though we could temporarily disable the check if @MikeInnes or anyone else with commit access don't help here
2. upper bound MacroTools 0.3.5 to be 0.4-only
3. lower bound MactoTools 0.3.5 to be 0.6-only

Option 1 would be easiest IMO. It may be possible to do both 2 and 3 but it would be the first time as far as I know that that union of version ranges feature was used in metadata, it's not a widely tested capability.