Enforce template syntax typename over class

[Repiteo]

Cleanup commit meant to standardize template class vs typename naming conventions, opting for the latter. Nothing functionally changes as a result of this, beyond uniformity across all templates.

[Mickeon]

Has this been agreed upon? Is typename favoured? I am to assume yes, but I wouldn't want the opposite to be true.

[AThousandShips]

I'm not convinced by this change, it's entirely aesthetic and clutters the log a lot

[Repiteo]

Has this been agreed upon? Is typename favoured? I am to assume yes, but I wouldn't want the opposite to be true.

...You know what, fair point. While it's strictly aesthetic regardless of what's chosen, I did jump the gun with a unilateral typename approach. I'll run this by #core to make sure there's no objections & make this a draft until a consensus is reached

I'm not convinced by this change, it's entirely aesthetic and clutters the log a lot

Shoot, thanks for reminding me about the log noise; added an accompanying .git-blame-ignore-revs change like the other aesthetic commits

[AThousandShips]

Also, this is really backwards IMO, there's far more cases of class than typename in the existing codebase, based on a cursory search (might not be complete but a search for class in template contexts yields 960 matches, and typename just 191)

[Mickeon]

Should we not have a chat about this on RocketChat? Have Akien chime in, too?

[Repiteo]

Agreed; made a post now in #core

[AThousandShips]

@jordo Please don't bump without contributing significant new information. Use the 👍 reaction button on the first post instead.

[aaronfranke]

Nothing functionally changes as a result of this

Are we sure about that? In MSVC, class and struct are different (beyond just default member visibility), so care should be taken to not mix them up. If we want these templates to be used with struct types, would typename be more agnostic for that? EDIT: Also, what if T is a primitive type like int or double?

[AThousandShips]

Replacing typename by class instead makes this far, far less disruptive, changing only 58 files and making 144 changes, as compared to 101 files with 538 changes

Also, I'd say typename hurts readability as it's hard to tell if it's a specifying a dependent template name

[akien-mga]

I don't care much either way personally. Both terms seem to be pretty much equivalent.

The main compelling arguments I've found online are:

1. Using typename removes the awkwardness of using class for types which aren't classes, such as primitive types.
2. Using class when dealing with actual classes, and typename with non-class types, can be a stylistic choice to convey meaning.

I think (1) is a decent argument for using typename everywhere, so it removes the awkwardness. It doesn't make things more informative on the other hand.

Going with (2) can be interesting, but requires some more work to keep consistent across the codebase than a plain "always using typename" guideline.

Also, I'd say typename hurts readability as it's hard to tell if it's a specifying a dependent template name

How so? What's the difference with class?

[Repiteo]

Are we sure about that? In MSVC, class and struct are different (beyond just default member visibility), so care should be taken to not mix them up. If we want these templates to be used with struct types, would typename be more agnostic for that? EDIT: Also, what if T is a primitive type like int or double?

From what I understand, templates handle the class keyword differently than class/struct are normally used; they're instead handled the way that most languages would use a Type keyword. That is to say, a template class parameter accepts classes, structs, and primitive types. I have no clue why they opted for this instead of typename exclusively—C++ can be frustratingly wishy-washy with syntax rules.

This perceived ambiguity is part of why I'm in favor of typename over class unilaterally. There's no ambiguity in a typename parameter accepting a struct type or primitive type, whereas a class parameter has a comparative chance (very small, but non-zero) of appearing like it's not suitable for accepting structs nor primatives as valid values.

[akien-mga]

Given the discussion so far, I'm in favor of the change, it's less ambiguous this way.

[akien-mga]

Thanks!

[dimkKy]

Nothing functionally changes as a result of this

Are we sure about that? In MSVC, class and struct are different (beyond just default member visibility), so care should be taken to not mix them up. If we want these templates to be used with struct types, would typename be more agnostic for that? EDIT: Also, what if T is a primitive type like int or double?

Starting from c++20 struct is a NTTP