Make Debug impls optional

[Ralith]

When disabled, this buys us a ~12% reduction in buildtime in my environment

I'd like some discussion on whether it should be enabled by default. It's less breaking that way, but I dunno how often people are going to scan the feature list for opportunities to improve buildtime.

[i509VCB]

I wonder should we list available features in the readme so people know the built time optimizations exist.

[Ralith]

Since nobody seems to feel strongly, I've updated this to disable the feature by default.

[MarijnS95]

I'm tracking the windows-rs repository, and users seem rather disgruntled by the removal of Debug. Not sure if they'd be as unhappy if there was a debug feature, even id disabled by default.

[Ralith]

I don't have enough experience with winapi to judge how the usage patterns might vary. I'd rather have this in but enabled by default than not in at all, if that's your preference? I'll ask around and see if anyone I know would be impacted; hard to predict in general.

[MarijnS95]

windows-rs, the replacement for winapi - gauging the few replies and thumbs up here, and the Rust API guidelines suggesting to implement it for public types: https://rust-lang.github.io/api-guidelines/debuggability.html#all-public-types-implement-debug-c-debug it seems better to leave it on by default? I'm still fine with having it as feature of course.

OTOH I'm not sure how useful our limited debugging implementation is: we don't even (and for good reason) dereference and follow pointers. Still, at least being able to see pointer values and everything else that's not a pointer already goes a long way. Still on the fence about default-enabling or disabling it...

[Ralith]

I think our current Debug impls are trivially useful for output structures, and potentially useful for input structures that might get handled by intermediate layers of abstractions in user code. Sure, a lot of stuff needs pointer context for completeness, but lots of stuff doesn't as well; e.g. being able to quickly and easily inspect memory barriers seems convenient.

Given the uncertainty, I think it's better for us to merge this with the feature enabled by default than agonize over it, so I've turned it back on. We can always disable in the future.

[MarijnS95]

@Ralith On my machine I can win another 400-500ms (7.9s down to 7.4s) by removing const_debugs: Ralith/ash@optional-debug...MaikKlein:optional-debug

We may want to preserve just the debug for Result though?

[Ralith]

I preserved Result (for unwrap()) and ObjectType (for debug messenger purposes).

[MarijnS95]

LGTM!

(I hope nobody notices the cfg_attr being below instead of above the derives like in vk/definition.rs - I did 😬)