Fix div and friends

[timholy]

Gracious me, coding by copy/paste does have its downsides.

This one was my fault from back in 2014, I believe. Amazing it's not been noticed before.

[vchuravy]

I am not on a computer right now so I couldn't just test this :)

Was the problem here the return type of the div operator? It seems odd to me that the result would be an integer (in all cases).

[timholy]

This is what happens on master:

julia> using FixedPointNumbers
INFO: Recompiling stale cache file /home/tim/.julia/lib/v0.5/FixedPointNumbers.ji for module FixedPointNumbers.

julia> x = 16uf8
UFixed{UInt8,8}(0.063)

julia> y = 2uf8
UFixed{UInt8,8}(0.008)

julia> div(x, y)
UFixed{UInt8,8}(0.031)

julia> div(Float32(x), Float32(y))
8.0f0

julia> r = 1uf8:1uf8:10uf8
UFixed{UInt8,8}(0.004):UFixed{UInt8,8}(0.004):UFixed{UInt8,8}(0.039)

julia> length(r)
------ InexactError -------------------- Stacktrace (most recent call last)

 [1] — length(::StepRange{FixedPointNumbers.UFixed{UInt8,8},FixedPointNumbers.UFixed{UInt8,8}}) at range.jl:364

 [2] — unsafe_length(::StepRange{FixedPointNumbers.UFixed{UInt8,8},FixedPointNumbers.UFixed{UInt8,8}}) at range.jl:361

InexactError()

None of those problems happen with this branch.

[vchuravy]

Hm I see. I am still trying to wrap my head around this. Isn't the problem
that the result of the division is not representable as a UF8? I am still
wondering how falling back to integer division solves this, wouldn't it
make more sense to fall back to floating point division?

Can I also note that the printing of 2uf8 is broken on master? Or the
conversion to Float32.

On Sun, 11 Sep 2016, 21:14 Tim Holy, [email protected] wrote:

This is what happens on master:

julia> using FixedPointNumbers
INFO: Recompiling stale cache file /home/tim/.julia/lib/v0.5/FixedPointNumbers.ji for module FixedPointNumbers.

julia> x = 16uf8
UFixed{UInt8,8}(0.063)

julia> y = 2uf8
UFixed{UInt8,8}(0.008)

julia> div(x, y)
UFixed{UInt8,8}(0.031)

julia> div(Float32(x), Float32(y))8.0f0

julia> r = 1uf8:1uf8:10uf8
UFixed{UInt8,8}(0.004):UFixed{UInt8,8}(0.004):UFixed{UInt8,8}(0.039)

julia> length(r)------ InexactError -------------------- Stacktrace (most recent call last)

[1] — length(::StepRange{FixedPointNumbers.UFixed{UInt8,8},FixedPointNumbers.UFixed{UInt8,8}}) at range.jl:364

[2] — unsafe_length(::StepRange{FixedPointNumbers.UFixed{UInt8,8},FixedPointNumbers.UFixed{UInt8,8}}) at range.jl:361
InexactError()

None of those problems happen with this branch.

—
You are receiving this because you commented.
Reply to this email directly, view it on GitHub
#47 (comment),
or mute the thread
https://github.com/notifications/unsubscribe-auth/AAI3aoPhk_fZTORDySMnW9r5KoX6YV6wks5qpKdXgaJpZM4J6Igf
.

[vchuravy]

This is what confuses me. The result type here is now an unsigned integer.

[timholy]

Are you arguing we should throw an InexactError?

[timholy]

Yes, it's not representable, but in general you expect it not to be. The definition of div is just "Computes x/y, truncated to an integer." So the result should be 8, not 0.031.

I'm not sure I follow what's broken about the printing/conversion to Float32:

julia> x = 2uf8
UFixed{UInt8,8}(0.008)

julia> Float32(x)
0.007843138f0

(except I'm bummed about how verbose the printing is, I thought we'd fixed that)

[vchuravy]

Ahhh, I am too tired. You are of course right.

• div should return integers.
• and I compared Float32(2uf8) to the wrong number

Thanks for explaining. LGTM from my side.