1
votes

This is related to JoGusto's answer at Casting Error: lvalue required as left operand of assignment. In the answer, he/she states:

but there is one case where it is not true: casting, then dereferencing a pointer:

*((int *) chrPtrValue) = some_integer_expression;

I think I found the answer at Joseph Mansfield answer from Why does an lvalue cast work?, where he cited the standard. But that confused me more because I can differentiate between lvalues and rvalues, but xvalues and prvalues are still new to me.

Naively, it seems to me that the rule is present for a reason, so some of the methods for circumventing it would probably be [indirectly or directly] illegal also.

I have a few questions related to the lvalue cast. The use cases includ the following. In the first case, the underlying types are different. In the second case, the qualifiers were changed.

float f;
*(static_cast<int*>(&f)) = 1;

int ptr = ...;
*(static_cast<volatile int*>(&ptr)) = NULL; 

Is it legal C and C++ circumvent the lvalue cast error using indirection then dereferencing?

If the cast only changes the qualifiers (i.e., static const or volatile) , then is it still legal in C and C++?

If its legal C and C++, then does it violate other rules, like GCC's aliasing rules?

Finally, if it does violate C or C++, or other rules, then what is the approved way to do it (perhaps a memcpy or memmove)?

2
There's no such thing as "C/C++", please pick one especially when talking about details like this. (And they're called aliasing rules, not anti-aliasing.)Mat
Thanks Mat. Informally, I was speaking to the C family of languages. C/C++ was changed to C and C++. But I also work in Objective C, so if you have insight there, then it would be appreciated, too. (And I fixed the title... The older I get, the more I transpose characters and words...)jww

2 Answers

1
votes

The JoGusto answer is (a) not very good, and (b) on a C question.

First, is it legal C and C++ circumvent the lvalue cast error like that (indirection then dereferencing)?

IDK what you mean by "circumvent the lvalue cast error". The code (T)x = y; is just illegal nonsense (except for the case in C++ where T is an lvalue reference, as Joseph Mansfield's answer covers). You don't circumvent it; you write code that has a sensible meaning and does what you want to do.

The code *(T *)ptr = y; compiles. It means to invoke the assignment operator on a T object which is stored at the address in ptr. It's the same as (T &)*ptr = y; in C++, which is reinterpret_cast<T&>(*ptr) = y; .

does it violate other rules, like GCC's anti-aliasing rules?

The behaviour is subject to alignment and strict aliasing. If there is not actually a T object stored at that address, nor an object of type compatible with T according to the list in the strict aliasing rule (in this case: int or unsigned int), then it is undefined behaviour.

then what is the approved way to do it (perhaps a memcpy or memmove)?

You could write:

int x = some_integer_expression;
memcpy(chrPtrValue, &x, sizeof x);

I can usually differentiate between lvalues and rvalues, and not xvalues and prvalues.

Which expressions are you having trouble identifying in this example?

0
votes

An "lvalue" cast is/was often used in writing compilers all the way up until 2010 or so. This loss of this extremely useful construct means that many legacy compilers such as pcom no longer compile in many environments.

Older versions of byacc -- perhaps as late as 1996 -- actually encouraged the use of this technique. If I remember correctly, the code in pcom used the technique extensively as did several pcc2-based compilers. (I know this from personal experience.)