prev_float(::BigFloat) is missing about intervalarithmetic.jl

[dpsanders]

Oops, thanks for catching this. I'll fix tomorrow

[dpsanders]

This just needs a definition of a fallback method
AdjacentFloats.prev_float(x) = prevfloat(x)

[kalmarek]

Thanks. Don't forget to define nextfloat as well ;-)

[dpsanders]

What context did you find this in? Can this be turned into a test case?
These should probably be defined directly in the AdjacentFloats package.

[JeffreySarnoff]

do you want me to add the fallback for BigFloats into AdjacentFloats?

[dpsanders]

Yes that would be great, thanks!

…

On 29 Jun 2017 3:47 p.m., "Jeffrey Sarnoff" ***@***.***> wrote: do you want me to add the fallback for BigFloats into AdjacentFloats? — You are receiving this because you commented. Reply to this email directly, view it on GitHub <#72 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AALtTm6m2V9OVRrXOeQUaWGWKV5etKk2ks5sI_9QgaJpZM4OEiSB> .

[JeffreySarnoff]

fixed in master, submitted as new release

[JeffreySarnoff]

merged

[kalmarek]

I will try to submit a test

[kalmarek]

Ok, test case for this turned out to be easier than expected:

using IntervalArithmetic
setrounding(Interval, :accurate)
w = 0 ± big(0.1)
w*w

(or any function that uses directed in rounding, e.g. :+, :-, :*, :/, :atan2, :^, :sqrt, :inv, :tanh, :asinh, :acosh, :atanh and all of CRlibm)
The catch is it happens only on :accurate rounding and julia-v0.5 does not recompile functions after running setrounding. So an eventual test:

• would have to start at the beginning of session
• set rounding before any of these are used
• run tests caching possible MethodError: no method matching prev_float(::BigFloat)

I didn't look much into tests for Intervals, but how feasible are those conditions?

[dpsanders]

0.5 support has been dropped in the latest version so that is no longer an issue.

It would be great if you could make a pull request with your suggested test code!

[kalmarek]

Should these be placed in IntervalArithmetic.jl/test/interval_tests/rounding.jl ?
since it's the definition of prevfloat, nextfloat should we test all of the functions above, or just + is enough?

[dpsanders]

Fixed upstream in latest release of AdjacentFloats.jl.