[likeliv]

I disagree that the user of the macro needs to understand its output. The output of a macro is an implementation detail, and the documentation of the macro should be enough to use the macro without even looking at its output.

For example, no need to understand the magic behind the 'quote!' or the #[async_trait] macros to use them.

[fauigerzigerk]

Not every user has to understand every macro. But the output of a Rust macro is valid Rust code whereas the output of a C compiler is not valid C code.

As a consequence, criticising the complexity of whatever a C compiler generates cannot possibly be valid criticism of C's complexity on a _semantic_ level whereas criticising the output of a Rust macro can be valid criticism of Rust on a semantic level.

[dtech]

Nearly all C compilers allow inline assembly. Macros are similar to inline assembly in that they step outside the normal bounds/use case of the language and are a complicated but valid and useful tool.

Most C programmers won't have to write or understand inline assembly often, if ever. Of course you can encounter it in production problem or something, so you could make an argument that all C programmers need to understand "C with inline assembly", which you are making for Rust macros.

As long as you just use Rust macro's and not write your own you are solidly in "C without inline assembly" territory.

[fauigerzigerk]

>Nearly all C compilers allow inline assembly. Macros are similar to inline assembly in that they step outside the normal bounds/use case of the language and are a complicated but valid and useful tool.

I couldn't disagree more. Macros are not similar to inline assembly at all precisely because they do _not_ step outside the bounds of the language.

Whatever similarities you may find, it's simply not helpful to deny the fundamental distinction between language A generating code in language A and language A invoking/generating code in language B.

It's futile to debate the properties of a particular language if you can't make a distinction between that language and anything it can generate or embed in some opaque way.

[dtech]

What I fail to understand is why it's so important to you whether a pre-processing/compiler/de-sugaring of language A results in a valid snippet of language A or another language for complexity of language A.

Take Objective-C automatic reference counting [1], implemented as a transformation of the original code to valid code of the same language (similar to a Lisp/Rust/Scala style macro) by automatically adding the appropriate statements.

If I understand your argument correctly, according to you this increases the complexity of "Objective-C with ARC", but would not have done so if the compiler would have implemented it as a direct transformation to its compilation target instead.

To me, that is an implementation detail which does not matter. "Objective-C with ARC" is exactly as complex in both cases. I'd argue it's even a little bit less complex with the "macro" implementation since you don't need to know assembly to know what ARC is doing.

Similar to ARC, Rust implements some things with macro's, which first "compiles" something to valid Rust. To me this is not more difficult for users than it would be if the compiler would directly generate LLVM IR without this intermediate step.

The inclusion of macro's in a language do make the language more complex of course! And creating macros is notoriously difficult since you're basically implementing a small compiler step! But for the user using something is not suddenly more difficult because it's implemented using a macro.

[1] https://en.wikipedia.org/wiki/Automatic_Reference_Counting

[fauigerzigerk]

>What I fail to understand is why it's so important to you whether a pre-processing/compiler/de-sugaring of language A results in a valid snippet of language A or another language for complexity of language A.

It's important because any and all code in language A is fair game when it comes to criticising semantic properties of language A. Code in other languages isn't.

noncoml criticised Rust based on a piece of Rust code. My point is simply that this criticism is potentially legitimate in ways that criticising C based on a piece of assembly code could never be.

I think our disagreement arises because you are asking a completely different question. What you're saying is that for devs who invoke some code it may not matter one bit whether that code was implemented in language A or language B or language A generated by language A or B. Those distinctions do not necessarily affect the semantic complexity for users of that code.

I completely agree with that. I also agree that the code snippet noncoml posted does not mean using async code in Rust has to be overly complicated.

But when I see a piece of Rust source code, I can criticise Rust based on it regardless of where that code came from or what purpose it serves.

Someone had to think in terms of Rust in order to write that code, and it's always worth asking whether it shouldn't be possible to express the same thing in a simpler way or whether that would have been possible in another language.

The fact that this code does not have to be understood by its users is completely irrelevant for this particular question.