Unexpected return value of NaN when raising to power

[milktrader]

julia> foo = 0.1; foo ^ .21
0.6165950018614822

julia> foo = -0.1; foo ^ .21
NaN

julia>  -0.1 ^ .21
-0.6165950018614822

[StefanKarpinski]

Well, that's no good.

[johnmyleswhite]

This doesn't happen on master anymore:

julia> foo = -0.1; foo ^ .21
NaN

The third case seems to be a parse ambiguity:

julia>  -0.1 ^ .21
-0.6165950018614822

This gets parsed as:

julia> -(0.1^.21)
-0.6165950018614822

[StefanKarpinski]

What version did you encounter this bug on?

[milktrader]

I haven't updated in a couple weeks ...

[milktrader]

julia> versioninfo()
Julia Version 0.3.0-prerelease+503
Commit cb64f5a* (2013-12-15 23:50 UTC)
Platform Info:
  System: Darwin (x86_64-apple-darwin13.0.0)
  WORD_SIZE: 64
  BLAS: libopenblas (USE64BITINT DYNAMIC_ARCH NO_AFFINITY)
  LAPACK: libopenblas
  LIBM: libopenlibm

[johnmyleswhite]

Maybe this was a very recent change:

julia> versioninfo()
Julia Version 0.3.0-prerelease+916
Commit a007350* (2014-01-11 01:19 UTC)
Platform Info:
  System: Darwin (x86_64-apple-darwin13.0.0)
  CPU: Intel(R) Core(TM) i7-3520M CPU @ 2.90GHz
  WORD_SIZE: 64
  BLAS: libopenblas (USE64BITINT DYNAMIC_ARCH NO_AFFINITY)
  LAPACK: libopenblas
  LIBM: libopenlibm

[milktrader]

Did an update just now

julia> foo = 0.1; foo ^ .21
0.6165950018614822

julia> foo = -0.1; foo ^ .21
ERROR: DomainError
 in ^ at math.jl:405

julia> -0.1 ^ .21
-0.6165950018614822

With this

julia> versioninfo()
Julia Version 0.3.0-prerelease+925
Commit bf8c1a2* (2014-01-11 15:39 UTC)
Platform Info:
  System: Darwin (x86_64-apple-darwin13.0.0)
  CPU: Intel(R) Core(TM)2 Duo CPU     P7350  @ 2.00GHz
  WORD_SIZE: 64
  BLAS: libopenblas (USE64BITINT DYNAMIC_ARCH NO_AFFINITY)
  LAPACK: libopenblas
  LIBM: libopenlibm

[johnmyleswhite]

I think that's the right behavior unless we want to change Julia's precedence rules.

[StefanKarpinski]

Yes, I think the current version is correct. Arguably the NaN was also right. I was misreading the example.

[jiahao]

#5234

[milktrader]

So when foo is a negative value, it gets parsed differently?

julia> foo = 0.1; foo ^ .21
0.6165950018614822

julia> foo = 0.1; -foo ^ .21
-0.6165950018614822

julia> foo = -0.1; foo ^ .21
ERROR: DomainError
 in ^ at math.jl:405

[StefanKarpinski]

Unary operators like - bind less tightly than ^.

[johnmyleswhite]

The relevant issue is that ^ binds tighter than -.

[StefanKarpinski]

It's very common to write things like -2x^2 or -x^2. These do not mean (-2x)^2 or (-x)^2 but rather -2(x^2) and -(x^2) respectively. This interpretation is completely standard in mathematical notation.

[milktrader]

Okay, I see this now. Thanks for the clarification. Sorry for the noise here.