Move icon toggling to a build script

[carloskiki]

No description provided.

[lpotthast]

Hi, could you sum up the implications of these changes? I also looked into fixing the feature problem today. Havent noticed the discussion two weeks ago. We really screwed crates.io with the amount of features 🙈..
The problem seems to be around the enums. Removing them allows proper dead code elimination. I thought that the same benefit, naming an icon without using the data struct, having something to serialize, etc., would also be possible by defining a unit struct per icon which could provide a &'static IconData. This would allow us to remove ALL features alltogether and users would also not have to name the icons they are using anymore. What do you think?

[lpotthast]

Ahh.. This would only work when dropping serialization support, as even a custom deserializer would bring in all known icon variants through a match statement... Might be somehing to reconsider.

[carloskiki]

Hi, could you sum up the implications of these changes? I also looked into fixing the feature problem today. Havent noticed the discussion two weeks ago. We really screwed crates.io with the amount of features 🙈..

Oh yes we did 😅

The problem seems to be around the enums. Removing them allows proper dead code elimination. I thought that the same benefit, naming an icon without using the data struct, having something to serialize, etc., would also be possible by defining a unit struct per icon which could provide a &'static IconData. This would allow us to remove ALL features alltogether and users would also not have to name the icons they are using anymore. What do you think?

Dead Code elimination already works with enum when using lto = true or lto = "fat" (from my testing). At first I wanted to simply remove all of the feature flags and constrain people to use lto as a "hot-fix" while we find a better solution. The only reason I decided to hack a build script together in the mean time is that importing every icon by default completely destroys rust-analyser as it pollutes the namespace.

I didn't think of the implications of serde and I think your solution is a better than mine for that reason. I haven't tested DCE with unit structs, but if it does work then I would be happy to implement your solution, but users would have to use the qualified syntax e.g., icondata::AiMyIcon if they want their IDE to work.