[jacquesm]

   int increment(a)

   {
      return a+1;
   }

   void somefunc()

   {
       int b = 5;

       b = increment(b);

       printf("b now %d\n", b);
   }

[everslick]

that exact example is given in the document and presented as 1 of 2 possible solutions. I believe, the critique in the parent comment was unwarranted, but I thought I maybe missed something.

[jacquesm]

This is the only proper way of doing that. Having pointers to variables on input that functions modify is ugly and error prone. C gets a lot better the moment you try to make your functions as pure as you can.

[wruza]

What do you think of

  T get_t(int how, char **error);
  int get_val(T *v);

One could rewrite these as

  struct get_t_res {T; char *};
  struct get_t_res get_t(int);

  struct get_val_res {int; T};
  struct get_val_res get_val();

But what do you think of it?

[jacquesm]

Less clear. I much prefer for functions to do what's on the tin in the most simple way possible.

[jcelerier]

As long as you don't return structs, as the compiler may introduce memcpys here. C doesn't have C++'s RVO, last time I checked many cases that GCC and Clang did optimize in C++ weren't optimized in C.

[pjmlp]

True, but still one way I appreciate making C safer is by using poor man's Abstract Data Types with translation units.

For the use cases that are too expensive to use functions, macros can be used instead.

However that is only if C++ cannot be used at all, otherwise don't bother.

[jacquesm]

That works, at the expense of possibly hiding bugs in plain sight. Macros are a double edged sword in that sense, you can almost but not quite create a DSL on top of C that is a lot more safe but it has its drawbacks, and you need to be very good at mentally modeling macro expansion to read and debug code like that in order to ensure that it does what it seems to be saying that it will do.

[pjmlp]

Agree, if it is me, I rather use inline functions, but anyway as mentioned only use C when there is no other option.

[tpush]

Do you have an example of that where C++ struct returns are optimized but not C ones?

[jcelerier]

enjoy: https://gcc.godbolt.org/z/7Pn8eqhdK

that's latest GCC, at -O3

[moonchild]

Clang does generate good code https://gcc.godbolt.org/z/z781r7Phh

But yes, sadly I think it's still generally advisable to avoid returning large structs by value.

Note that this has nothing to do with RVO; as I understand it, RVO is about eliding copy constructors, not actual memory copies, and c has no copy constructors and so gets 'RVO' in all cases.