Replace manual enum functions with autogenerated ones

[MarijnS95]

This is the gtk-rs counterpart of https://gitlab.freedesktop.org/gstreamer/gstreamer-rs/-/merge_requests/630, following the introduction of autogenerated associated function support in gtk-rs/gir#991.

Turns out that, as far as I was able to run the tests (after extracting them from CI.yml...), the only functions that were defined by hand are in pango's Gravity enum 🎉
(testing with --all-features doesn't work because that inadvertently removes functions through cfg(not()) which are used by examples)

Note that this includes gtk-rs/gir#994 to get rid of doc(cfg()) on match arms and expressions. Solves a ton of warnings (When compiling with dox) but at the same convolutes this PR. See the second commit for an actually relevant change.

Let's hold off merging until gir is ready.

[sdroege]

Looks generally good to me, thanks!

[MarijnS95]

@sdroege Pushed!

Turns out I have more write!(f, ...) -> f.write_str(...) changes lying around but that results in a whopping 400 files changed. I don't think that's worth taking in assuming format strings like "{}" or literals without formatting arguments are optimized out?

[sdroege]

I don't think that's worth taking in assuming format strings like "{}" or literals without formatting arguments are optimized out?

I don't think that assumption is necessarily correct :) But better put those into a separate PR, not as part of this one. That's an easy to review change.

[MarijnS95]

@sdroege Hence the question mark, I'm curious yet have never looked into the resulting assembly from all these formatting macros. I assumed it to be pretty efficient but that might have been massively wrong :)

[sdroege]

@sdroege Hence the question mark, I'm curious yet have never looked into the resulting assembly from all these formatting macros. I assumed it to be pretty efficient but that might have been massively wrong :)

Well, IMHO it's also nicer to read code-wise so definitely a useful change even if it compiles to the same code. It actually might as the format string is handled at compile-time.

[MarijnS95]

Finally got around to writing the three variants in godbolt: https://godbolt.org/z/v3xoxK

Note that this is might be hard to relate as to_str is inlined and we don't have an external C function call the compiler knows nothing about. But then again we're all about the formatting, let's see:

write!(f, "{}", self.to_str()) is the worst. Besides write!() expanding to .write_fmt which needs to set up an std::fmt::Arguments object, the pointer to <str as Display>::fmt is pushed into this object resulting in virtual/dynamic dispatch when formatting arguments are iterated. The virtual function results in a .write_str. Note that all this happens on the same formatter/Write instance, it's not like things are copied around multiple times :)

Next is writing a literal with write!(f, "My value"). This might be used in a match self { EnumVariant => write!(f, "Enum variant"), }. As above, this has to set up std::fmt::Arguments for .write_fmt.

Finally there is .write_str(). This is what .write_fmt uses internally so it's obvious we just got rid of all the extra work around formatting, which must inevitably be faster.
In addition it's easier to use in self.write_str(self.to_str()) and self.write_str(match self { EnumVariant => "Enum variant", }) becomes more elegant.

[sdroege]

Ok, so as expected I guess? :)

[MarijnS95]

@sdroege Yup. It's not all too suboptimal. The function pointer Display::fmt for &str is pushed into the Arguments object which is called from write_fmt. The implementation simply calls .pad() which expands to a plain write_str when there are no padding options in the current formatter. We'd be getting rid of the virtual functions and extra boilerplate, not much else besides cosmetics. Don't think that's a change worth making.

[sdroege]

Otherwise looks good

[sdroege]

@MarijnS95 Your latest gir changes don't affect this one here? Otherwise please also update it :)

[MarijnS95]

@sdroege Nope, all clear here :)

[sdroege]

@MarijnS95 Please update this one with the latest gir, and then let's get it in :)

[MarijnS95]

@sdroege Thanks again! Updated with the latest hashes :)