Windows- or 32bit-specific InexactError

[timholy]

Appveyor builds are failing:
https://ci.appveyor.com/project/timholy/images-jl/build/1.0.83/job/sgy1gmtq36rbsemh
https://ci.appveyor.com/project/timholy/images-jl/build/1.0.80/job/wv5yw4qxj0fvugqw
on the following command:

using FixedPointNumbers
aa = Ufixed12[0.6 0.2;
              1.4 0.8]

Can anyone on Windows test this? I don't understand what the problem could be.

[elsuizo]

In my 32-bits Debian laptop with Julia 0.3.7 works fine Tim

[timholy]

Thanks for the useful data point!

[timholy]

Ah, but it will probably only fail on 0.4---trunc didn't throw errors on 0.3.

[tkelman]

Ufixed12[0.6 0.2; 1.4 0.8] in isolation doesn't error for me on win32, 0.4, but it does give different answers every time I run it.

Pkg.test("Images") fails for me on win32, 0.4, with the same error you're seeing on appveyor.

[timholy]

That's alarming. Sounds like some kind of memory error?

I'm going to file this at FixedPointNumbers and CC some folks who know more. Just to double-check, this works for you on Linux, right? (You have Linux available, right?)

[timholy]

Sorry to pester you more, but can you also test things like:

Ufixed12(0.2)
Ufixed12(1.4)
Ufixed12([0.2,1.4])

and let me know what happens?

[tkelman]

Just to double-check, this works for you on Linux, right? (You have Linux available, right?)

Yes and yes.

Ufixed(0.2)
Ufixed(1.4)
Ufixed([0.2,1.4])

MethodError: rawtype` has no method matching rawtype(::Type{FixedPointNumbers.Ufixed})``

With Ufixed12(0.2) things appear to work. Worth noting that for the array Ufixed12[0.6 0.2; 1.4 0.8] it looks like the first row is always zeros, the second row has different values every time.

[timholy]

Sorry for initially omitting the 12, and thanks for catching it.

[timholy]

I think I'm going to file this in julia proper, since I don't think this can be limited to FixedPointNumbers.

[tkelman]

Some more info: it's a 32 bit issue, but with different symptoms between Linux vs Windows:

  | | |_| | | | (_| |  |  Version 0.4.0-dev+4524 (2015-04-27 00:18 UTC)
 _/ |\__'_|_|_|\__'_|  |  Commit 982c60a* (0 days old master)
|__/                   |  i686-linux-gnu

julia> using FixedPointNumbers

julia> Ufixed12[0.6 0.2; 1.4 0.8]
ERROR: InexactError()
 in trunc at float.jl:341
 in hvcat_fill at abstractarray.jl:824
 in typed_hvcat at abstractarray.jl:843

[timholy]

That makes more sense than it being OS-specific!

[timholy]

Ah, above I see that it might have worked on 0.3.7. You don't happen to have a 32bit (either Windows or Linux) to test on with 0.3, do you?

[tkelman]

Looks like the bug is not present on 0.3.8-pre on win32.

[timholy]

Thanks, that's useful to confirm.

[johnnychen94]

Since this issue is five years old and the current CI passes, I'm closing it now.