fix coverage test on nightly

[simeonschaub]

Depends on JuliaLang/julia#42810 to pass. I did also observe some other
test failures on nightly I couldn't really explain.

[codecov]

Codecov Report

Merging #498 (2231cda) into master (f365bcd) will increase coverage by 0.60%.
The diff coverage is n/a.

@@            Coverage Diff             @@
##           master     #498      +/-   ##
==========================================
+ Coverage   86.13%   86.73%   +0.60%     
==========================================
  Files          12       12              
  Lines        2394     2488      +94     
==========================================
+ Hits         2062     2158      +96     
+ Misses        332      330       -2     

Impacted Files	Coverage Δ
src/precompile.jl	0.00% <0.00%> (ø)
src/utils.jl	86.07% <0.00%> (+0.04%)	⬆️
src/builtins.jl	76.64% <0.00%> (+0.08%)	⬆️
src/optimize.jl	97.53% <0.00%> (+0.14%)	⬆️
src/localmethtable.jl	96.61% <0.00%> (+0.18%)	⬆️
src/construct.jl	91.26% <0.00%> (+0.18%)	⬆️
src/breakpoints.jl	95.10% <0.00%> (+0.25%)	⬆️
src/commands.jl	93.33% <0.00%> (+0.27%)	⬆️
src/packagedef.jl	91.48% <0.00%> (+0.58%)	⬆️
src/types.jl	74.52% <0.00%> (+0.83%)	⬆️
... and 1 more

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update f365bcd...2231cda. Read the comment docs.

[KristofferC]

I wonder how the resolver error on 1.0 started happening.

[simeonschaub]

That's my fault. I added DiffUtils as a test dependency to make future debugging easier, but forgot that it requires Julia >= 1.3

[timholy]

Ooh, that's unfortunate. I don't immediately understand why it's different from the old heuristic, but I haven't poked at it. Can you file a Julia issue? I regard line-number issues as bugs.

[aviatesk]

Is this really bug? The line numbers appearing in lowered code seems just identical:

v1.7

julia> ex = Main, Base.parse_input_line("""
           x = 1
           for i = 1:10
               x += 1
           end
           let y = 0
               z = 5
           end
           if 2 > 1
               println("Hello, world!")
           end
           @elapsed sum(rand(5))
           """; filename="fake.jl"); Meta.lower(Main, ex)
:($(Expr(:toplevel, :(#= fake.jl:1 =#), :(x = 1), :(#= fake.jl:2 =#), :(for i = 1:10
      #= fake.jl:3 =#
      x += 1
  end), :(#= fake.jl:5 =#), :(let y = 0
      #= fake.jl:6 =#
      z = 5
  end), :(#= fake.jl:8 =#), :(if 2 > 1
      #= fake.jl:9 =#
      println("Hello, world!")
  end), :(#= fake.jl:11 =#), :($(Expr(Symbol("hygienic-scope"), quote
    #= timing.jl:297 =#
    while false
        #= timing.jl:297 =#
    end
    #= timing.jl:298 =#
    local t0 = time_ns()
    #= timing.jl:299 =#
    $(Expr(:escape, :(sum(rand(5)))))
    #= timing.jl:300 =#
    (time_ns() - t0) / 1.0e9
end, Base))))))

nightly

julia> ex = Base.parse_input_line("""
           x = 1
           for i = 1:10
               x += 1
           end
           let y = 0
               z = 5
           end
           if 2 > 1
               println("Hello, world!")
           end
           @elapsed sum(rand(5))
           """; filename="fake.jl"); Meta.lower(Main, ex)
:($(Expr(:toplevel, :(#= fake.jl:1 =#), :(x = 1), :(#= fake.jl:2 =#), :(for i = 1:10
      #= fake.jl:3 =#
      x += 1
      #= fake.jl:4 =#
  end), :(#= fake.jl:5 =#), :(let y = 0
      #= fake.jl:6 =#
      z = 5
  end), :(#= fake.jl:8 =#), :(if 2 > 1
      #= fake.jl:9 =#
      println("Hello, world!")
  end), :(#= fake.jl:11 =#), :($(Expr(Symbol("hygienic-scope"), quote
    #= timing.jl:356 =#
    $(Expr(Symbol("hygienic-scope"), :($(Expr(:meta, :force_compile))), Base.Experimental))
    #= timing.jl:357 =#
    local t0 = time_ns()
    #= timing.jl:358 =#
    $(Expr(:escape, :(sum(rand(5)))))
    #= timing.jl:359 =#
    (time_ns() - t0) / 1.0e9
end, Base))))))

The actual difference seems to be that v1.7 produces NewvarNode while nightly doesn't:

v1.7

julia> modexs = collect(ExprSplitter(Main, ex)); frame = Frame(modexs[5]...); frame.framecode.src
CodeInfo(
    @ fake.jl:11 within `top-level scope`
1 ─     Core.NewvarNode(:(#310#t0))
   ┌ @ timing.jl:297 within `macro expansion`
2 ┄│     goto #4 if not false
3 ─│     goto #2
   └
   ┌ @ timing.jl:298 within `macro expansion`
4 ─│     _J1 = ($(QuoteNode(time_ns)))()
│  └
│  ┌ @ timing.jl:299 within `macro expansion`
│  │     rand(5)
│  │     sum(%J5)
│  └
│  ┌ @ timing.jl:300 within `macro expansion`
│  │     ($(QuoteNode(time_ns)))()
│  │     ($(QuoteNode(-)))(%J7, _J1)
│  │     ($(QuoteNode(/)))(%J8, 1.0e9)
│  └
└──     return %J9
)

nightly

julia> modexs = collect(ExprSplitter(Main, ex)); frame = Frame(modexs[5]...); frame.framecode.src
CodeInfo(
    @ fake.jl:11 within `top-level scope`
   ┌ @ timing.jl:356 within `macro expansion`
1 ─│     $(Expr(:meta, :force_compile))
│  │ @ timing.jl:357 within `macro expansion`
│  │     _J1 = ($(QuoteNode(time_ns)))()
│  │ @ timing.jl:358 within `macro expansion`
│  │     rand(5)
│  │     sum(%J3)
│  │ @ timing.jl:359 within `macro expansion`
│  │     ($(QuoteNode(time_ns)))()
│  │     ($(QuoteNode(-)))(%J5, _J1)
│  │     ($(QuoteNode(/)))(%J6, 1.0e9)
│  └
└──     return %J7
)

So I'd say the first line number of that chunk was fake.jl sort of accidentally.

[timholy]

xref #500

[timholy]

You sure you need this? It didn't fail for me on nightly.

[aviatesk]

Yeah, I couldn't reproduce the error within REPL, but I also confirmed Pkg.test will hit this line consistently without this change (maybe because the compilation heuristic used during Pkg.test is different from that within REPL?).
I think the root problem here is that @interpret doesn't configure enter_generated when calling enter_call_expr.

[timholy]

I'm OK with this if it's necessary.

[timholy]

I think one of the failures should be fixed by #500 instead, but otherwise I'm good with this.

[timholy]

Sorry for the rebase. Once that's done let's merge this.