I'm more of the mindset that writing something like this is probably a code smell to begin with. Is there any reason I'm not thinking of right now, that this couldn't be typedef'd and refactored into something far more readable?
C gets a lot of blame for pointer gibberish like this but quite honestly you can write gibberish in any language. I don't see any fundamental or technical reason you couldn't write clean, readable C.
Why do people continue to use this term "code smell", instead of "hard to read code" or something similar, more equivalent??? First seem almost offensive to an author.
“Code smell” does not mean “hard to read code”, it means “an unreliable but still useful indication that there may be something wrong in the design of the code connected to the piece described as having the ‘smell’”.
And the reason it continues to be used is that it is a concise idiom that is useful.
"an unreliable but still useful indication that there may be something wrong in the design of the code"
Well, but you just proved the point - it can be described in non offensive manner. And more over, judging something based on gut feeling ("indication") may be actually even worse, as you may offend someone who did actually a good job.
Sorry if "nit picking", recently was reading a lot about burn outs in the industry, and this is a thing that did catch my attention...
I don't keep up with the latest C shenanigans, cuz I like C the way it was, but have they change something where "pointer to an array[5]" is a meaningful distinction to draw?
I mean, "a pointer to an array of X" is simply "a pointer to X" and using hungarian notation, you can encode the knowledge "this pointer can be incremented" into its name.
and typedef'ing *function declarations? who has families of functions with the same type signatures that they want to point to?
> but have they change something where "pointer to an array[5]" is a meaningful distinction to draw
This has been a meaningful distinction since at least C99.
> "a pointer to an array of X" is simply "a pointer to X"
They aren't actually the same. The former can be decayed into the latter but they aren't actually equivalent and you'll get type mismatches if you try to treat them as the same thing.
> who has families of functions with the same type signatures that they want to point to
An example is callbacks or handlers for responding to interrupt requests. A lot of hardware interface code relies on typedef-ed function decls because very often you are passing user side functions through your interface so that you can stash those function pointers somewhere and invoke them when some event occurs.
that's an array of 3 pointers to ints. if you pass foo as an argument you get a pointer to a pointer to an int (with knowledge if you can hang onto it that there are more pointers to ints lined up in memory)
The trick is to just assert equal types. Most compilers have extensions that allow you to easily compare type equality (and C23 actually standardizes the `typeof()` operator).
So you basically take your ugly type, put it in a #define and then create a static assertion that matches the type against said ugly type.
Now the compiler will throw a shit fit if the types don't match. Have fun breaking the type up into smaller pieces until you have something legible.
var x: pointer to func() pointer to array[5] of pointer to func() char;
or if you wish to replace some keywords with glyphs:
var x: ^func() ^[5] ^func() char;
And it's always a nice puzzle for the reader to explain why there are three "pointer" in cdecl output and three carets in the ALGOL-like declaration, but only two asterisks in the C declaration.
In this case, the C declaration doesn't match the other two. The variable x is a function that returns a pointer to an array of 5 pointers to functions returning char. Indeed, that's what cdecl.org says:
declare x as function returning pointer to array 5 of pointer to function returning char
Using the notation you did, that would be:
var x: func() ^[5] ^func() char
There are only two arrows there now.
If you wanted a pointer to a function like this, you would need a third asterisk in the declaration:
Oh, good catch, thank you! But I remember an example with some other tricky C expression/type declarator where the number of actual dereferences differed from the amount of asterisks in the code.
> Using the notation you did, that would be:
Well, it'd be
func x() ^[5] ^func() char; ...
because it's a function declaration, after all, not a variable.
Rust isn't so bad, is it? The example of `char ((x[3])())[5]` would translate to `[fn() -> [fn() -> u8; 5]; 3]`. It's inherently an ugly type, but I think it's easier to read than the C version.
I was thinking of the issue of going from English to C. Even when you out in malformed C it's nice to get either a compiler error which explains the problem, or even better an LLM that can suggest how to fix it.
I don't think it is gibberish. It's code and in order to read that code you need to understand the language, and to understand language you need learning and experience.
Maybe it can be useful for learning, but if you have to use such tool, I suspect you won't understand it anyway - so in a way it is more a gibberish-to-gibberish translator.
I disagree. The conventions for declaring arrays, pointers, and function pointers are all idiosyncratic. In C, the type is always to the left of the variable being declared. Except for arrays, which have part of the declaration to the right. And except for pointers, which need to be affixed to every item if there are multiple declarations. And except for function pointers where you need to wrap the variable name like (*name). Individually I can wrap my head around these exceptions, but putting all of them together, it's just hard to read.
It takes a pretty smart person to do that. Which is pretty confusing.
How can such a smart person not not understand how all things that are possible are not all equally good?
The fact that both the compiler and you can parse that doesn't make it a good way to document or convey meaning or intent.
C is chock full of inconsistencies and ambiguities that are only disambiguated by essentially being a human compiler and maintaining the same parsing state-machine manually in your head to know what any given "(" actually means or does. As a self-proclaimed fluent C linguist, you know this better than most.
All coding involves that of course but all implimentations are not equally unhelpful.
The cpu and some people can read the binary itself. They just need to know the cpu's opcodes, documented right in the datasheet that anyone can read.
C gets a lot of blame for pointer gibberish like this but quite honestly you can write gibberish in any language. I don't see any fundamental or technical reason you couldn't write clean, readable C.
And the reason it continues to be used is that it is a concise idiom that is useful.
Well, but you just proved the point - it can be described in non offensive manner. And more over, judging something based on gut feeling ("indication") may be actually even worse, as you may offend someone who did actually a good job.
Sorry if "nit picking", recently was reading a lot about burn outs in the industry, and this is a thing that did catch my attention...
I mean, "a pointer to an array of X" is simply "a pointer to X" and using hungarian notation, you can encode the knowledge "this pointer can be incremented" into its name.
and typedef'ing *function declarations? who has families of functions with the same type signatures that they want to point to?
This has been a meaningful distinction since at least C99.
> "a pointer to an array of X" is simply "a pointer to X"
They aren't actually the same. The former can be decayed into the latter but they aren't actually equivalent and you'll get type mismatches if you try to treat them as the same thing.
> who has families of functions with the same type signatures that they want to point to
An example is callbacks or handlers for responding to interrupt requests. A lot of hardware interface code relies on typedef-ed function decls because very often you are passing user side functions through your interface so that you can stash those function pointers somewhere and invoke them when some event occurs.
I don't believe these two are the same. An "array of X" indeed decays to a "pointer to X". But a "pointer to an array of X" is something else. E.g.
Perhaps the first two are what you mean, though, and this is just a terminology issue.Granted, the function pointer syntax is forever confusing (to me anyway). The rest is easily tackled by naming things.
Even for function pointers, it’s just one lookup and then you can copy-paste the typedef for any other function pointer types in the project.
(NOT the author. It simply helped me.)
What if you're given somebody else's code and you need to understand it to put a typedef there
So you basically take your ugly type, put it in a #define and then create a static assertion that matches the type against said ugly type.
Now the compiler will throw a shit fit if the types don't match. Have fun breaking the type up into smaller pieces until you have something legible.
(Yes, this is a joke)
[1] https://docs.racket-lang.org/htdp-langs/beginner.html
Then again people complain that they are too verbose, and they rather write in hieroglyph friendly languages.
If you wanted a pointer to a function like this, you would need a third asterisk in the declaration:
> Using the notation you did, that would be:
Well, it'd be
because it's a function declaration, after all, not a variable.Any language with type after name :
Next I want one to explain some of Rust’s more cryptic pointer gibberish. Usually I just hit “use suggested fix X” until the compiler’s happy.
cdecl is always correct with regards to this stuff.
I don't know why you'd choose the former.
Output for the example I got on opening the website:
cdecl.org: declare x as function returning pointer to array 5 of pointer to function returning charChatGPT: x is a function that, when called, gives us access to 5 functions that each return a character. (TL;DR, it gave a full explanation too)
Like mentioned before the error rate of LLMs is probably much higher on complex expressions.
I don't think it is gibberish. It's code and in order to read that code you need to understand the language, and to understand language you need learning and experience.
Maybe it can be useful for learning, but if you have to use such tool, I suspect you won't understand it anyway - so in a way it is more a gibberish-to-gibberish translator.
How can such a smart person not not understand how all things that are possible are not all equally good?
The fact that both the compiler and you can parse that doesn't make it a good way to document or convey meaning or intent.
C is chock full of inconsistencies and ambiguities that are only disambiguated by essentially being a human compiler and maintaining the same parsing state-machine manually in your head to know what any given "(" actually means or does. As a self-proclaimed fluent C linguist, you know this better than most.
All coding involves that of course but all implimentations are not equally unhelpful.
The cpu and some people can read the binary itself. They just need to know the cpu's opcodes, documented right in the datasheet that anyone can read.
typedef uint64_t qbb_t __attribute__((vector_size(sizeof(uint64_t) * 4)))
Syntax error
OK, its an extension, meh.