Add feature tags to signify engine float precision

[mihe]

This adds double (edit: and single) as a "feature tag" to OS::has_feature to signify whether an engine was built with double-precision, also known as float=64 or REAL_T_IS_DOUBLE.

This allows, for example, the engine to load different extension libraries (see here) depending on if the engine was built with float=32 or float=64, allowing extension authors to bundle libraries for both precision types.

Without this ability to differentiate between float=32 and float=64 the engine will crash when loading an extension with a mismatching precision type.

[mihe]

It's worth mentioning that this PR will be further complemented by #67906, which will make the engine prioritize the more specific entry in a .gdextension file, for cases where there are multiple matching entries, like so:

[libraries]
linux.x86_64.double = "res://my-extension-double-precision-x64.so"
linux.x86_64 = "res://my-extension-single-precision-x64.so"

It's also worth mentioning that this PR does not address the case where a float=64 engine tries to load an extension which does not explicitly specify a double library, like so:

[libraries]
linux.x86_64 = "res://my-extension-single-precision-x64.so"

... which would match both float=32 and float=64, causing the engine to load it either way and subsequently crash.

This could be partially addressed by also adding a single feature tag, or however you want to name it, to signify single-precision (float=32), allowing extension authors to be specific about that condition as well, like so:

[libraries]
linux.x86_64.single = "res://my-extension-single-precision-x64.so"

Let me know if you think something like single should be added to this PR as well.

[JoNax97]

I think double_precision or float_64 would be clearer names

[mihe]

Fair point! Something like float_64 would make the naming of the single-precision equivalent a bit less weird as well, with float_32. I'll go ahead and change that right away.

[mihe]

There, I've renamed double to float_64 and went ahead and added float_32 as well.

[Mickeon]

Hold on. #67399 has been approved (not merged as of the time writing this). So "double" may make more sense.

[mihe]

Ah, yes, thank you for linking that PR.

In that case I agree that something like double/single would make more sense.

I guess another thing to consider with regards to the naming is that the editor already adds .double as a suffix to its executable when you build it with double-precision, so there is some precedence there.

Also, if I understand it correctly, when #67906 ends up being merged, extensions will be able to use their library filenames as feature tags, without specifying anything in their .gdextension file. So having my-extension.double.x86_64.dll (which lines up with the editor executable) would check for the features double and x86_64.

I'll leave it unchanged for now, for further discussion.

[mihe]

Considering the 4.x milestone this was just given I went ahead and changed the names back to the more likely double/single.

[aaronfranke]

Having feature tags like this is a good idea, and the specific names proposed here are very good names.

[mihe]

Any chance of having this merged for 4.0?

[Calinou]

Looks good to me. After this is merged, remember to update the Feature tags documentation as well (you can open a PR as draft right away).

[mihe]

you can open a PR as draft right away

Done: godotengine/godot-docs#6476

[akien-mga]

Thanks!