Use a macro to set the DefaultPrecision

[elvisdukaj]

DefaultPrecision is fixed to double, making it hard for this library to work with other frameworks using different precision. This PR allows to define the precision with a macro, and use double as default fallback for not breaking existing code.

[edrosten]

I'm a little reluctant to introduce this macro, since it semi globally changes the meaning of code. You can still use any precision by doing, e.g. Vector<2, float>. In addition you could add since template aliases such as template<int I> using fVector=TooN::Vector<I, float>; Can you give an example of where it would be helpful? I could offer some suggestions, I've used toon without modification on floats and even once modular integers.

…

-Ed

On Wed, Jun 12, 2024, 13:41 Elvis Dukaj ***@***.***> wrote: DefaultPrecision is fixed to double, making it hard for this library to work with other frameworks using different precision. This PR allows to define the precision with a macro, and use double as default fallback for not breaking existing code. ------------------------------ You can view, comment on, or merge this pull request online at: #18 Commit Summary - 793418d <793418d> Use a macro to set the DefaultPrecision File Changes (1 file <https://github.com/edrosten/TooN/pull/18/files>) - *M* TooN/TooN.h <https://github.com/edrosten/TooN/pull/18/files#diff-2c0ce493689b3967db4551c08ec52ce79c9011870edbc9052772b3e8ae6395ac> (6) Patch Links: - https://github.com/edrosten/TooN/pull/18.patch - https://github.com/edrosten/TooN/pull/18.diff — Reply to this email directly, view it on GitHub <#18>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAIZN2Y5BBQ63K3PBG7QEHLZHAXYLAVCNFSM6AAAAABJGFWIYOVHI2DSMVQWIX3LMV43ASLTON2WKOZSGM2DQNJVGY2DGNQ> . You are receiving this because you are subscribed to this thread.Message ID: ***@***.***>

[elvisdukaj]

Thank you very much for the quick feedback! I am porting some old code to conan2 and internally we are"hacking" the source code to redefine typedef float DefaultPrecision. This value is used internally, and the code fails to build with the upstream version. I don't know the rationale of changing this value to float, but I can investigate.

I can leave with the hack for now, I applied a patch to the conan recipe to make it work like before.

I'm a little reluctant to introduce this macro, since it semi globally changes the meaning of code

We are already using this in production in multiple OSs (Windows, Linux, Android, macOS, iOS) and multiple architectures (x86_64, arm8). The default value anyway won't be changed but it will give the possibility to customize this from outside. Do you see something risky to do?

[edrosten]

Thinking about it, I do still think it's something that can be quite dangerous to use, and can set one up for trouble. On the other hand, sometime you need escape hatch (see the FORTRAN one below).