supporting float like fncls

[vianneylanguille]

Hi !

Do you plan to integrate floating point values support in supported dtypes ? Like in fncls

Thanks

Vianney

[endrebak]

That would be a much simpler thing to implement in this library. I will do it.

This should also be part of the RangeFrame (the class PyRanges inherits from like IRanges in GenomicRanges). It should support floating point operations.

[endrebak]

I think in src/ruranges.structs::5 we can just add whatever the float trait is called here:

pub trait PositionType: PrimInt + Signed + Hash + Copy + radsort::Key + Element + Copy + PartialOrd + ToPrimitive + Zero + std::fmt::Display {}
impl<T> PositionType for T where T: PrimInt + Signed + Hash + Copy + radsort::Key + Element + Copy + PartialOrd + ToPrimitive + Zero + std::fmt::Display {}

@marco-mariotti Related question: do you mind if PyRanges also support floating point starts/ends? It does not make biological sense but it would take more work to restrict the types to only be ints. We could warn the user in the summary string under the. pyranges __repr__.

[vianneylanguille]

Thanks !

And side question again, do you think it's feasible to publish a windows-compatible version on conda ? Currently there is linux/mac versions available, but part of my work is on windows. And there is no 'as fast' alternative available on windows (pure python only..). I can't use pip version as my company does not allow much stuff to be installed such as compilers on windows

Regards

Vianney

[endrebak]

You can try to make a pr. I have no way to test it nor the time. But I think it should be perfectly feasible. I'm not sure any change is needed to the actual code in my repository. It's my be that pyoxide supports Windows perfectly.

[endrebak]

I'm not opposed to any such change. It seems like the setup files in this Project might need some minor tweaks.

[vianneylanguille]

ok, I created a PR on bioconda recipes to add win-64 in the platforms.

[vianneylanguille]

PR is waiting for review, but I read somewhere that bioconda does not build for windows. So if they confirm that, it won't be possible to have a condz package for windows

[endrebak]

You could try building the package on your own computer and ask to use the .dll or what it would be on your work computer?

[vianneylanguille]

Yes, but the plan was to replace, in a package lots of people as using at work (50% on linux and 50% on windows), an interval intersection computation that is currently pure numpy (vectorized, thus use a lots of memory) by the one that is in ruranges, while keeping the overall dependencies workflow as 'neat' as possible. An alternative would be to host the recipes of ruranges on conda-forge instead of bioconda since conda forge offers windows build capabilities, but this introduce some change for people currently using it. Vianney Le sam. 8 nov. 2025, 15:51, endrebak ***@***.***> a écrit :

…

*endrebak* left a comment (pyranges/ruranges_py#16) <#16 (comment)> You could try building the package on your own computer and ask to use the .dll or what it would be on your work computer? — Reply to this email directly, view it on GitHub <#16 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ACA7NNTRZF76USB4DEZ4JM333X7PJAVCNFSM6AAAAACLL6WCPSVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZTKMBWGYYTAMRZGE> . You are receiving this because you authored the thread.Message ID: ***@***.***>

[endrebak]

This is next on my list of things to do. Sorry for being so late to follow this up but my vacation did not start until now.

[endrebak]

Can I just throw if I encounter a NaN?

[vianneylanguille]

Hi, Yes, this should not be a problem. Besides, bioconda replied to my PR, they don't build on windows. Regards Vianney Le dim. 30 nov. 2025, 21:27, endrebak ***@***.***> a écrit :

…

*endrebak* left a comment (pyranges/ruranges_py#16) <#16 (comment)> Can I just throw if I encounter a NaN? — Reply to this email directly, view it on GitHub <#16 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ACA7NNTD4UD2H76UQQLVY5D37NHJ7AVCNFSM6AAAAACLL6WCPSVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZTKOJTGMYDCMZRG4> . You are receiving this because you authored the thread.Message ID: ***@***.***>

[endrebak]

I will take a break working on this.

[endrebak]

Are you fine with IEEE exact float compare? So that the numbers would need to have the exact same representation to be considered equal? This means that stuff like the below can happen

a_end = 0.1 + 0.2       # 0.30000000000000004
b_start = 0.3           # 0.29999999999999999 (conceptually)
print(a_end == b_start) # False