> This construct works perfectly fine in C

Intuitively, I would say that this is actually undefined behavior (it would probably be difficult to expose a wrong behavior in practice though).

In C specs, I found 6.5.2.2, paragraph 9:

> If the function is defined with a type that is not compatible with the type (of the expression) pointed to by the expression that denotes the called function, the behavior is undefined.

We might discuss whether

    void (*)(char *)

is "compatible" with

    void (*)(void *)

but I think it isn't, since:

    void target(void *ptr) {}
    void (*name)(char *ptr) = target;

fails to compile with the error message:

    initialization of ‘void (*)(void *)’ from incompatible pointer type ‘void (*)(char *)’

The compiler explicitly says "incompatible pointer type".

Same for:

    void target(char *ptr) {}
    void (*name)(void *ptr) = target;

    void (*name)(char *ptr);
    typedef void (*name_func)(char *ptr);

    void target(void *ptr)
    {
       printf("Input %p\n", ptr);
    }


    char *data = "string";
    name = (name_func)target; // Illegal: casting fn that takes void* to a fn that takes char*
    name(&data); // Illegal: passing a char** into a function that takes char*

  /* declarations */
  void qsort(void *lineptr[], int left, int right,
             int (*comp)(void *, void *));
  int numcmp(char *, char *);

  /* the offending line */
  qsort((void **) lineptr, 0, nlines-1, 
    (int (*)(void*,void*)(numeric ? numcmp : strcmp));

This is one of the very rare cases in C where something is technically Undefined Behaviour, but in practice works and is recommended.

The typedef struct trick, is very common idiom that creates _more_ safety, not less. All reasonable compilers should (and do) support this. It is sad that the ISO standard is not in line with reality at all times. (I say that as a member of the wg14 and the UB study group)

I Recommend Daniel keep his typedef struct definition, and then have an ifdef to revert to the void definition for when Clang does its UB sanitizer. While checking for prototype discrepancies is a very good thing to automate, Clang should add an exception for this.

Awesome writeup. Always interesting to read what Daniel has to say.

I think the fact that it turned out that he was wrong (and UBsan was right, as usual) is a great testament to the shortcomings of C.

Lots of people - both inexperienced and very experienced - celebrate it for being "simple" and "close to the hardware", but the truth of the matter is that it is precisely not close enough to the hardware for people who _know_ what the hardware is doing to be able to do what they expect, and it's too close to the hardware to be able to be able to ignore it.

Lots of experienced C programmers (and - guilt by association - C++ programmers as well) run into UB because they have clear expectations of the compiler. I.e., they know what the compiler should generate, more or less, and C is just a convenient notation. But compilers don't live up to those expectations, because they don't actually compile your code for the hardware. They compile it to the virtual machine abstraction defined by the standard, which very often works differently from any real architecture, and then translate that into machine code. Even though there is basically a single set of semantics that every single "relevant" (mainstream) architecture implements. This is a holdover from when C had to target architectures that are 100% irrelevant today.

Everybody's favorite example is signed integer overflow. In both x86-64 and ARM64, that just works - two's complement is the only relevant implementation, so there's no issue. But `int` in C and C++ is not that.

Almost every single common UB pitfall has reasonable behavior at the assembler level for every mainstream architecture, and almost every single niche architecture.

C gives you the illusion of being close to the hardware, but in actual reality the hardware is several steps removed, so if you want to leverage your knowledge of the hardware, calling conventions, assembly, or other low-level details, you have to go out of your way to work around the C standard.

(Aside: We need new languages to tackle this, and I coincidentally happen to like Rust. Lots of people coming from C or C++ are irritated and frustrated by Rust, but 99% of the time it's because Rust gives you a compile error where C would give you UB. This is one example of that out of thousands.)

I would think the proper way to do this would be

  #if defined(BUILDING_LIBCURL)
    struct Curl_easy {
      …
    }
  #else
    struct Curl_easy;
  #endif
  typedef struct Curl_easy CURL;

If BUILDING_LIBCURL isn’t defined that tells code “CURL is identical to a struct named Curl_easy”. If it is defined, that also tells code what fields it has.

Interesting problem. The typical solution in C++ to deal with type erasing function pointer types is to go through a trampoline function:

    struct X {};
    void use_x(X*);
    using F = void(void*);
    void bar(F* fn, void* y) { fn(y); }

    template<class T, auto fn>
    void trampoline(void*arg) { return fn(reinterpret_cast<T*>(arg)); }

    X x;
    bar(&trampoline<X, use_x>, &x);

In plain C there is no way to generate the trampoline at the point of use in the same way template instantiation works, but it can be generated by a macro at global scope.

Yes, it's very much undefined behavior. As I recall, GTK's glib does this all over the place for signals.

edit: I'm not advocating that this being UB is fine. I don't expect compilers to exploit this for optimization, because so many projects rely on this working. There might be room to extend compatible function types to make this defined.

> In 2016 I wanted to change the type universally to just typedef struct Curl_easy CURL; … as I thought we could do that without breaking neither API nor ABI.

This seems to be the obvious solution and how most libraries define their opaque handles. It doesn't break a guaranteed API and it doesn't break the ABI any more than than only using it when building the library - and you can check that it doesn't break the ABI on any platform where you want to guarantee ABI stability.

Ok, this is UB, calling a function pointer through a different type than its definition is a pretty clear example of UB. The problem here is that there's a confusion between "void * and char * are implicitly convertible in C" and "void * and char * are the same". The latter is true for many platforms (especially older ones) but not all (I think there were platforms where functionally they had `typedef char void`). There's a side note of conflating "this has defined behavior on my platform that is stable and works for me" and "it's not UB if it's stable and works on a platform", just like integer overflow is UB despite being entirely defined behavior on every platform under the sun.

Anyway, if folk are curious there are many platforms where not only can the representation of `void()(void)` and `void()(char)` be different - even if pointing to the same function - but the representation of even just the data pointers void* and char* may not be the same, again while pointing to the same memory.

For example, on platforms with pointer authentication function pointers are generally (I would say "always" but in principle it can be avoided) signed, and in some configuration the type of the function is incorporated into the function. Calling the function pointer requires authenticating the pointer, and authenticating the pointer requires that the call site agrees on the type of the pointer because otherwise the signature fails.

Absent actual pointer auth hardware you could imagine someone implementing this as some kind of monstrosity like this (very hypothetical, unpleasant, and footgun heavy) horror:

    #define SIGN_FPTR(fptr) (typeof(fptr))(((uintptr_t)fptr)|(MAGIC_HASH(stringify(typeof(fptr)) << some_number_of_bits))
    #define AUTH_FPTR(fptr) (typeof(fptr))(((uintptr_t)fptr)^(MAGIC_HASH(stringify(typeof(fptr)) << some_number_of_bits))

and you can immediately see that if you had code that used these but disagreed on the type of the function you'd have a bad time. With compiler+hardware pointer auth this is just handled transparently.

In principle a pointer auth environment could apply this type discrimination logic to data pointers as well, but I'm unaware of any that do so implicitly. But if a platform did do so, then the incorrect type of the parameter would mean you would fail inside the function, if you were able to call it (say if you weren't using a function pointer, but had mistyped the prototype).

Similarly, in other environments, the pointer may be directly aware the type being referenced, and I believe that CHERI supports this, in which case even if you could call the function pointer when attempting to read the pointer I believe it would fail.

Having got here, you might be saying "but hang on C says void* and char* are the same", and we go all the way back to my first sentence where I said "are implicitly convertible" :D

On plenty of systems casting from one pointer type to another is an entirely source level feature and once lowered is completely invisible. But in the environments we're discussing

   (Type1*)pointerToType2

Under pointer auth it requires re-signing the pointer (you have to auth the original value to verify it, and then sign the result according to the new schema), and under CHERI I believe there are instructions for controlling how a pointer is tagged.

But the important thing is the C does not say they are the same thing, just that the conversion is automatic, just like numbers and bools, or numbers and bools in JS, or numbers and strings in JS, or objects and strings in JS, or nothing and strings in JS, or .... :D

Man rants about not expecting weird type system abuse that works on his machine to be undefined behavior

They say Rust is much harder than C for... disallowing these kinds of things?

There is no problem with using the typedef struct trick. The problem is assuming that such pointers have the same underlying representation as void pointers. Structs typically having stricter alignment requirements (whereas void* must be able to point to any byte) is one reason why their representation may differ from void*. This is why only [[un]signed] char* is guaranteed to have the same representation as void*.

Given:

    typedef struct foo Foo:
    Foo * createFoo();
    void useFoo(Foo * foo);

    Foo * foo = createFoo();
    void * vp = foo;

This is perfectly fine:

    useFoo((Foo *) vp);

This is not:

    void (*)(void *) vf = (void (*)(void *)) useFoo;
    (*fv)(vp);

> I say that as a member of the wg14 and the UB study group

Since you have experience in this area, do you know how likely it is that something like this could be resolved? I.e., if someone proposed "just make this defined", how likely would the C standards body be to agree and do so?

> very rare cases in C where something is technically Undefined Behaviour, but in practice works and is recommended.

In the bad old days, such cases were very common. I think complaining about it is a part of the process that slowly resolves it.

Many of such issue were also fixed in languages like Ada and Modula-2, among others, which C and C++ folks used to call "programming with a straighjacket".

Many folks will complain about anything that breaks their beloved illusion that C is a fancier macro assembler, except even proper macro assemblers have less UB than regular C.

Being close to the hardware isn't a dogma. The point of being close to the hardware is to be able to write code that is (nearly) as perfomant as it could be if you used assembler. Undefined behavior that let's the compiler better reason about your code like signed integer overflow is entirely compatible with that goal. If anything, unsigned overflow should also be UB unless you explicitly tell the compiler that an operation should wrap, which really is not something you want in the common case where any overflow already means you have a bug.

Wouldn't a faster and easier solution for existing code be to come up with a standard set of behaviours based on common architectures and make that a flag that can be switched on so expectations can be met (and call it "expected-C" or for the pun "ocean")?

C does give a lot of low-level control, but it is still an abstraction - as it should be.

Signed integer overflow is something were I personally like that it is UB, because I can turn on UBSan and find overflow bugs. If it were defined to wrap-around I could not do this.

Does anyone know of languages that achieve this? I'm interested, I'm currently implementing a project in x86 assembly for this reason, and am happy to try a higher level language.

I think this is more of an edge case than you give it credit for.

(External linkage) functions, and their callsites, are quite special as they straddle the boundary between actual machine and virtual machine.

The calling convention makes fairly strict requirements of what that must compile down to, since caller and callee could be dynamically linked objects compiled by different compilers or even entirely different languages.

If you define a function that takes a void* parameter it will read a pointer from the RDI(?). If you pass a char* it will pass the pointer in that same register.

Now it must be said that if the compiler can see both the definition and call, then it's free to do whatever it wants e.g. inline or something.

But yeah it's a bit complicated?

For signed overflow it's fascinating, Herb Sutter (WG21 convenor, Microsoft employee) writes that checking would "incur unacceptable costs".

Now, C++ programmers, were you consulted about these costs? Herb insists they're "unacceptable" but he provides no further information as to what the cost actually was, or who decided whether that cost was acceptable, much less how this could generalize across a wide variety of domains and platforms.

What's Herb's answer? You might hope that Herb would say OK, we'll provide wrapping for these types by default, it's not checked arithmetic but at least it's not UB. Nope.

You are confusing "the same" with "compatible" and maybe also with "have the same alignment and representation".

This is not a rant, but a well laid-out description of something a widely used software ran into as a result of a change in the compiler they use.

The #ifdef they had in place was a bit hackish, but I wouldn't call it abuse. It is a typical construction for C to do stuff like that which in general isn't without risk, but they have used it without any problems for years. The whole point of the blog post is to start a discussion whether sth. like this should rightfully be flagged as "unexpected behavior" or not.

And apparently worked the last 28 or so years on loads of other machines (I think the question would be where doesn't curl run)?

It's literally the opposite of "works on my machine" because it's in curl (which is most likely the single most widely deployed code base in the world).

How would you implement objects and inheritance in C without function casts? CPython certainly uses these casts.

dlysm() even relies on a (void *) cast that is not C standard compliant.

C is useless for certain applications with this "undefined behavior".