Reproduction of hashing test failure

[GunnarFarneback]

For me, hashing is an X rather than a ☠️. Anyway, this is as far as I can reduce it. Probably not minimal but rather brittle.

With /tmp/hashing_test.jl containing

vals = Any[
    [0], [1], [2],
    [2, 1, 2, 3], [2, 3, 2, 1], [2, 2, 3, 2],
    [0, 0], [0, 0, 0], [0, 1], [1, 0],
    [0, 0, 1], [0, 1, 0], [1, 0, 0], [0, 1, 2],
    [0, 0], [1, 0], [0, 1], [0, 0], [0, 0],
    [5, 1], [1, 1], [0, 2], [0, 2], [4, 0],
    [0, 0], [1, 0], [0, 0, 2], [0 0 7],
    [4 0 0], [0 2 4],
]

b = vals

then

include("utils.jl")
dotest1("hashing", "/tmp/hashing_test.jl", 10000)

results in

ERROR: UndefVarError: vals not defined
Stacktrace:
 [1] _step_expr!(::Array{JuliaStackFrame,1}, ::JuliaStackFrame, ::Any, ::JuliaProgramCounter, ::Bool) at /home/gunnar/.julia/dev/JuliaInterpreter/src/interpret.jl:419
 [2] limited_finish_and_return!(::Array{JuliaStackFrame,1}, ::JuliaStackFrame, ::Int64, ::JuliaProgramCounter, ::Bool) at /home/gunnar/.julia/dev/JuliaInterpreter/test/utils.jl:197
 [3] (::getfield(Main, Symbol("##11#13")){Int64})(::JuliaStackFrame) at /home/gunnar/.julia/dev/JuliaInterpreter/test/utils.jl:210
 [4] #invokelatest#1 at ./essentials.jl:761 [inlined]
 [5] invokelatest at ./essentials.jl:760 [inlined]
 [6] lower!(::Any, ::Dict{Module,Array{Expr,1}}, ::Array{LineNumberNode,1}, ::Module, ::Expr) at /home/gunnar/.julia/dev/JuliaInterpreter/test/utils.jl:96
 [7] lower_incrementally!(::Any, ::Dict{Module,Array{Expr,1}}, ::Array{LineNumberNode,1}, ::Expr, ::Module, ::Expr) at /home/gunnar/.julia/dev/JuliaInterpreter/test/utils.jl:74
 [8] lower_incrementally!(::Any, ::Dict{Module,Array{Expr,1}}, ::Array{LineNumberNode,1}, ::Expr, ::Module, ::Array{Any,1}) at /home/gunnar/.julia/dev/JuliaInterpreter/test/utils.jl:83
 [9] lower_incrementally!(::Any, ::Dict{Module,Array{Expr,1}}, ::Array{LineNumberNode,1}, ::Expr, ::Module, ::Expr) at /home/gunnar/.julia/dev/JuliaInterpreter/test/utils.jl:47
 [10] lower_incrementally! at /home/gunnar/.julia/dev/JuliaInterpreter/test/utils.jl:40 [inlined]
 [11] lower_incrementally at /home/gunnar/.julia/dev/JuliaInterpreter/test/utils.jl:36 [inlined]
 [12] (::getfield(Main, Symbol("##10#12")){String,Int64,Module,Expr})() at /home/gunnar/.julia/dev/JuliaInterpreter/test/utils.jl:296
 [13] cd(::getfield(Main, Symbol("##10#12")){String,Int64,Module,Expr}, ::String) at ./file.jl:96
 [14] dotest1(::String, ::String, ::Int64) at /home/gunnar/.julia/dev/JuliaInterpreter/test/utils.jl:293
 [15] top-level scope at REPL[2]:1

Removing e.g. the first element from vals makes the test pass.

[GunnarFarneback]

It passes when I test it on Julia 1.1. The above was with master.

[GunnarFarneback]

Running the hashing tests on Julia 1.1 fails quite differently from Julia master. Are both interesting to pin down or is the focus on master?

[timholy]

Interesting! What happens here is that lower_incrementally splits this into two chunks. On some Julia versions, the first chunk takes more than 10^4 statements to complete; if it aborts, then vals doesn't get defined and consequently the second expression errors.

This makes it clear that the nstmts limit is not as straightforward a solution as I had hoped. In principle it would be possible to detect this kind of interdependency automatically by scanning the lowered code for rhs Symbols and then seeing if they are already defined in JuliaTests. We could add a 2nd category of failure for this.

[timholy]

Table in #13 updated, now we are much closer.