Reverse Dependency Backwards Compatability Testing

[ripytide]

I've tested some of the most popular reverse dependencies of rgb to test the main branch's backwards compatibility:

• inferno: Compiles but some tests fail due to formatting differences, not a big deal as inferno can re-generate it's test results but we could also revert our formatting back to no-spaces.

  
  If you have modified the code and specifically want to update the expected test output, set `INFERNO_BLESS_TESTS=1` before the `cargo test`.

Diff < left / right > : < Flame Graph

Flame Graph

• image: Compiles and all tests pass.
• gif-dispose: Compiles and all tests pass.
• ravif: Compiles and all tests pass, gives deprecation warnings about types being renamed.
• resvg: Compiles, has no tests.
• femtovg: Compiles, fails the same tests with and without patching with the new rgb branch so seems like an unrelated bug or user-error on my part.
• lodepng: Compiles, fails the same test with and without patching with the new rgb branch so seems like an unrelated bug or user-error on my part.
• csscolorparser + rust-rgb feature: Compiles and all tests pass, gives deprecation warnings about types being renamed.
• ansi_colours: Bingo! Spits out two types of compiler errors when using the new rgb branch:

  
  error[E0308]: mismatched types
  --> /home/ripytide/.cargo/registry/src/index.crates.io-6f17d22bba15001f/empfindung-0.2.6/src/to_lab_impls.rs:45:57
  |
  45 |     fn to_lab(&self) -> (f32, f32, f32) { lab_from_grey(**self) }
  |                                           ------------- ^^^^^^ expected `u8`, found `Gray_v09<u8>`
  |                                           |
  |                                           arguments to this function are incorrect
  |
  = note: expected type `u8`
          found struct `Gray_v09<u8>`
  note: function defined here
  --> /home/ripytide/.cargo/registry/src/index.crates.io-6f17d22bba15001f/empfindung-0.2.6/src/to_lab_impls.rs:55:4
  |
  55 | fn lab_from_grey(grey: u8) -> (f32, f32, f32) {
  |    ^^^^^^^^^^^^^ --------

error[E0599]: no method named into_u8 found for reference &Gray<C> in the current scope --> src/impls.rs:78:36	78	fn as_u32(&self) -> u32 { self.into_u8() as u32 * 0x010101 }	^^^^^^^

= help: items from traits can only be used if the trait is implemented and in scope note: Component defines an item into_u8, perhaps you need to implement it --> src/impls.rs:28:1	28	trait Component: Copy {	^^^^^^^^^^^^^^^^^^^^^ help: one of the expressions' fields has a method of the same name
78	fn as_u32(&self) -> u32 { self.0.into_u8() as u32 * 0x010101 }
++

help: one of the expressions' fields has a method of the same name	78	fn as_u32(&self) -> u32 { self.v.into_u8() as u32 * 0x010101 }	++ help: there is a method into with a similar name
78	fn as_u32(&self) -> u32 { self.into() as u32 * 0x010101 }

The root cause seems to be the lack of a `Deref<Target=T>` for the `Gray_v08` types since we made it implement `Deref<Target=Gray_v09>` instead, we might have to revert back to the old `Deref` implementation.

- `i-slint-core`: Compiles, fails the same tests with and without patching with the new `rgb` branch as I don't have one of the test fonts installed.

[kornelski]

The mss_saliency crate didn't work without the legacy From impls, because it depends on RGB<u32>: From<RGB<u8>>.

https://github.com/kornelski/mss_saliency/blob/2e685c5f5c247d7e73c261b2fb46f02c08b2e7fe/src/lib.rs#L18;L23

[kornelski]

use rgb::*; use rgb::alt::*; GRAY8 breaks, because it's defined in two places differently.

[ripytide]

Those two issues seem like they would require major rework to fix, is it worth the effort to fix those backwards incompatibilities or can we simply expect users to fix those issues themselves after we release v0.8.90?

I think we should release v0.8.90 as is, given we've attempted backwards compatibility on a best-effort basis, how about you?

[kornelski]

I've removed spaces from Display.

[kornelski]

Unfortunately Rust can't display deprecations for uses of Deref, so moving users away from Gray's deref will be tricky.

[ripytide]

Hopefully the Deref issue won't be that bad as eventually Gray_v08 will be deprecated so they can migrate away from Deref at the same time they migrate to Gray_v09.

[kornelski]

Now it's impossible to future-proof access to Gray due to the .0 field and lack of Deref. Maybe both types should get .v() or .value() getter?

There's also no Rgb::from(gray)