What causes this code to work only on some compilers (UD?)?

Question

In my code I had this line:

f(t->a, t->b, destroy(t));

Where f is a function, t is a pointer to a struct and destroy is a function which frees the pointer t and returns NULL. The code compiled and run successfully on some compilers (cc, gcc, clang on OS X and clang on Linux) as I expected - t->a and t->b were evaluated before freeing t. However, when compiling with cc/gcc on Linux I got a segmentation fault, as if I was trying to dereference null pointer. What causes that?

Thanks!

haccks · Answer 1 · 2015-04-28T16:20:46.343

4

Order of evaluation of arguments of a function call is uspecified. destroy(t) may evaluate first and t will be free and therefore t->a will invoke undefined behavior.

C11: 6.5.2.2 Function calls (p12):

In the function call
(*pf[f1()]) (f2(), f3() + f4())
the functions f1, f2, f3, and f4 may be called in any order. [...]

edited Apr 28 '15 at 16:20

answered Apr 28 '15 at 16:09

haccks

104,019
25
176
264

In other words, the order depends completely on the compiler? – Dunno Apr 28 '15 at 16:10
1

In theory, a compiler could have various order of arguments evaluation for different call sites of the same function. – Basile Starynkevitch Apr 28 '15 at 16:19

What causes this code to work only on some compilers (UD?)?

1 Answers1

C11: 6.5.2.2 Function calls (p12):