search

kornelski / rust-rgb

struct RGB for sharing pixels between crates
https://lib.rs/rgb
MIT License
98 stars 19 forks source link

Add type aliases for pixel types #119

Closed ripytide closed 4 months ago

ripytide commented 4 months ago

From #64:

Type hints are repetitive, and other forms require turbofish.

Some more questions might be whether we provide these aliases for all pixel types or just Rgb and Rgba? and for how many integer types, u8, u16, isize? And from #25 whether we go with Rgb8 or RgbU8?

kornelski commented 4 months ago

For all pixel types, for consistency. I haven't seen usize/isize used. It probably doesn't appear outside of temporary calculations like pixel.map(), so it doesn't need an alias.

Just u8 would probably cover majority of uses. With u16 and f32, probably 99%.

But the naming is such a pain. RGBA8 is super common, and an old established name, everywhere outside of Rust. RgbaU8 is so unusual, that Google ignores the spelling, and DDG finds Rust crates (probably this discussion soon!)

And to make it worse, new GPU apis and shader languages have chosen to use a suffix instead, so there's RGBA32F.

So I'm torn between following Rust's naming, and having it weird for everyone else, or having a mishmash of "standard", but non-Rust names.

ripytide commented 4 months ago

u8, u16, f32 and f64 sound good to me for aliases.

The specific naming is indeed a pain.

My one strong opinion in on RGBA vs Rgba since the latter is much nicer to use when inside a rust code-base since it looks more like a type and less like a constant.

Consistency tells me more explicit is better so RgbaU8/Rgba8U is better than Rgba8 to prevent potential future collisions such as between RgbaF16 and RgbaU16.

But I have very weak feelings about postfix vs prefix for the type letter: RgbaU8 vs Rgba8U, in fact I think I prefer the postfix for floating point but the prefix for unsigned integers which seems super odd and inconsistent: RgbaU8 and Rgba32F.

kornelski commented 4 months ago

Having slept on it, I think I'll keep the RGBA8 style aliases, and add RGBA32F. This reduces code churn for existing users, and gives a choice between rusty Rgba::<u8> and industry-wide naming schemes.

ripytide commented 4 months ago

That sounds good, strikes a balance between the two approaches.

ripytide commented 4 months ago

This seems fixed in 30db416