Tests failing in julia nightly

[waldyrious]

See for example https://travis-ci.org/JuliaMath/Tau.jl/jobs/245339601 (from #25):

INFO: Testing Tau
Error During Test
  Test threw an exception of type ErrorException
  Expression: #= /home/travis/.julia/v0.7/Tau/test/runtests.jl:39 =# @inferred(sintau(T(x))) == sign(x) * zero(T)
  return type BigFloat does not match inferred return type Any
  Stacktrace:
   [1] macro expansion at /home/travis/.julia/v0.7/Tau/test/runtests.jl:39 [inlined]
   [2] anonymous at ./<missing>:?
   [3] include_from_node1(::Module, ::String) at ./loading.jl:549
   [4] include(::Module, ::String) at ./sysimg.jl:14
   [5] process_options(::Base.JLOptions) at ./client.jl:310
   [6] _start() at ./client.jl:378
ERROR: LoadError: There was an error during testing
while loading /home/travis/.julia/v0.7/Tau/test/runtests.jl, in expression starting on line 38

[giordano]

The problem is with T = BigFloat, but I can't understand why inference fails here.

Anyway, as I said in PR #19, sintau and costau can be improved by adapting them to new definitions of sinpi and cospi, which are more efficient.

[waldyrious]

Do you have plans to work on those anytime soon? Tau day is coming up (a week from now), and I'd like to make a release for the occasion -- it would be nice if we could include this fix in it. What are the main hurdles to adjust sintau and costau to match the new implementations of sinpi/cospi from Base, in your view?

[giordano]

No, I don't plan to work on this, mostly because internals of sinpi and cospi are obscure to me, so I wouldn't know how to adapt to this case :)

[waldyrious]

Yeah, they look pretty opaque to me too. I'll probably spend some time "decoding" them once I get the time.

That said, do you think there's a way we could adjust the current code to avoid this failure? We can add a separate issue to track the re-syncing of those methods' code with the corresponding versions in Base.

[giordano]

To avoid the failure we can just remove the @inferred, but the test will fail also at

Tau.jl/test/runtests.jl

Line 97 in a9142e3

@test Base.return_types(f,Tuple{T}) == [T]

when T = BigFloat.

Maybe try asking on discourse if someone has a clue about why sintau is unstable with T = BigFloat input?

[giordano]

I think this is no more an issue:

julia> using Tau, Test

julia> @inferred sintau(big(6.28))
0.9822872507286883882901415948346156477307161222321188530528350428954190590918724

@waldyrious could you please confirm?

[devmotion]

Works for me as well.