JETAnalyzer: false positive errors that can be analyzed to be safe by concrete evaluation

[Krastanov]

On Julia nightly from today:

v = Vector{Union{UInt8,Int,Float64}}()
push!(v, 0x0)
push!(v, 1)
push!(v, 1.0)
function f(vec)
    vectail = vec[2:end]
end
rep = @report_call f(v)

reports

═════ 1 possible error found ═════
┌ @ REPL[7]:2 vec[2 : lastindex(vec)]
│┌ @ array.jl:925 copyto!(X, firstindex(X), A, first(I), lI)
││┌ @ array.jl:319 Base._copyto_impl!(dest, doffs, src, soffs, n)
│││┌ @ array.jl:327 unsafe_copyto!(dest, doffs, src, soffs, n)
││││┌ @ array.jl:281 srcp = pointer(src, soffs)
│││││┌ @ abstractarray.jl:1243 Base._memory_offset(x, i)
││││││┌ @ abstractarray.jl:1247 Base.elsize(x)
│││││││┌ @ abstractarray.jl:251 Base.elsize(Base.typeof(A))
││││││││┌ @ array.jl:213 Base.aligned_sizeof(T)
│││││││││┌ @ reflection.jl:415 al = Base.datatype_alignment(T)
││││││││││ no matching method found `datatype_alignment(::Type{Union{Float64, Int64, UInt8}})`: al = Base.datatype_alignment(T::Type{Union{Float64, Int64, UInt8}})
│││││││││└─────────────────────

Versions:

julia> versioninfo()
Julia Version 1.10.0-DEV.193
Commit 162ee48e1c3 (2022-12-25 12:36 UTC)
Platform Info:
  OS: Linux (x86_64-linux-gnu)
  CPU: 8 × Intel(R) Core(TM) i7-10510U CPU @ 1.80GHz
  WORD_SIZE: 64
  LIBM: libopenlibm
  LLVM: libLLVM-14.0.6 (ORCJIT, skylake)
  Threads: 4 on 8 virtual cores

Julia nightly installed with unreleased juliaup (patched to support nightlies)

JET version v0.6.21

Should I be reporting things like this on the JET issue tracker or on the Julia one?

[jakobnissen]

I don't think this is an issue with JET, however, it may provide a option for improving JET. Here is a reduced MWE:

julia> using JET

julia> f(::DataType) = false
f (generic function with 1 method)

julia> g(T::Type) = Base.isbitsunion(T) ? true : f(T)
g (generic function with 1 method)

julia> @report_call g(Union{Int, UInt})
═════ 1 possible error found ═════
┌ @ REPL[3]:1 f(T)
│ no matching method found `f(::Type{Union{Int64, UInt64}})`: f(T::Type{Union{Int64, UInt64}})
└─────────────

Julia is not able to statically infer that Union{Int, UInt} is Base.isbitsunion due to a @nospecialize in the definition of the latter. Therefore, it's not able to statically infer that a Union will never hit f.

The code in your OP example is intentionally marked as @nospecialize, presumably for latency reasons.
JET could avoid this false positive if it ignored @nospecialize - but then, this might make JET latency absolutely terrible.

[lgeissbauer-btig]

This seems simular

┌ @ C:\MyPkg.jl:35 result = MyPkg.DataFrame()
│┌ @ C:\.julia\packages\DataFrames\dgZn3\src\dataframe\dataframe.jl:304 DataFrames.:(var"#DataFrame#207")(pairs(NamedTuple()), #self#)
││┌ @ C:\.julia\packages\DataFrames\dgZn3\src\dataframe\dataframe.jl:306 DataFrame(Base.vect(), DataFrames.Index())
│││┌ @ C:\.julia\packages\DataFrames\dgZn3\src\dataframe\dataframe.jl:187 [quote](true, #ctor-self#, columns, colindex)
││││┌ @ C:\.julia\packages\DataFrames\dgZn3\src\dataframe\dataframe.jl:229 DataFrames.convert(Vector{AbstractVector}, columns)
│││││┌ @ array.jl:613 T(a)
││││││┌ @ array.jl:621 Base.copyto_axcheck!(Vector{AbstractVector}(Base.undef, size(x)), x)
│││││││┌ @ abstractarray.jl:1180 copyto!(dest, src)
││││││││┌ @ array.jl:339 copyto!(dest, 1, src, 1, length(src))
│││││││││┌ @ array.jl:314 Base._copyto_impl!(dest, doffs, src, soffs, n)
││││││││││┌ @ array.jl:327 unsafe_copyto!(dest, doffs, src, soffs, n)
│││││││││││┌ @ array.jl:304 Base._unsafe_copyto!(dest, doffs, src, soffs, n)
││││││││││││┌ @ array.jl:245 destp = pointer(dest, doffs)
│││││││││││││┌ @ abstractarray.jl:1243 Base._memory_offset(x, i)
││││││││││││││┌ @ abstractarray.jl:1247 Base.elsize(x)
│││││││││││││││┌ @ abstractarray.jl:251 Base.elsize(Base.typeof(A))
││││││││││││││││┌ @ array.jl:213 Base.aligned_sizeof(T)
│││││││││││││││││┌ @ reflection.jl:415 al = Base.datatype_alignment(T)
││││││││││││││││││ no matching method found `datatype_alignment(::Type{AbstractVector})`: al = Base.datatype_alignment(T::Type{AbstractVector})

│││││││││┌ @ C:\file.jl:188 requests = Base.pop!(unique_symbol_2_requests, unique_symbol, Markets._Synchronization.nothing)
││││││││││┌ @ dict.jl:620 Base._pop!(h, index)
│││││││││││┌ @ dict.jl:585 Base._delete!(h, index)
││││││││││││┌ @ dict.jl:636 Base._unsetindex!(h.keys, index)
│││││││││││││┌ @ array.jl:178 pointer(A, i)
││││││││││││││┌ @ abstractarray.jl:1243 Base._memory_offset(x, i)
│││││││││││││││┌ @ abstractarray.jl:1247 Base.elsize(x)
││││││││││││││││┌ @ abstractarray.jl:251 Base.elsize(Base.typeof(A))
│││││││││││││││││┌ @ array.jl:213 Base.aligned_sizeof(T)
││││││││││││││││││┌ @ reflection.jl:415 al = Base.datatype_alignment(T)
│││││││││││││││││││ no matching method found `datatype_alignment(::Type{SharedDataTypes.Securities.UniqueSymbol})`: al = Base.datatype_alignment(T::Type{SharedDataTypes.Securities.UniqueSymbol})

[jakobnissen]

This issue appears solved on Julia master and the latest version of JET:

julia> @report_call g(Union{Int, UInt})
No errors detected

[aviatesk]

Yeah, #529 fixed this problem. Thanks for the reminder.