Undefined behaviour with #define in C

Question

I've got a define in C, looks like this

#define ROW_SIZE ID_SIZE + USERNAME_SIZE + EMAIL_SIZE

It evaluates to 293 Then I do this

uint32_t num_rows = pager->file_length / ROW_SIZE;

pager->file_length is 0 for sure
The result of num_rows is 289. Even 0 / ROW_SIZE evaluates to 289.

Why is the answer to expression wrong?

`#define ROW_SIZE ID_SIZE + USERNAME_SIZE + EMAIL_SIZE` -> `#define (ROW_SIZE ID_SIZE + USERNAME_SIZE + EMAIL_SIZE)` — Jabberwocky, Aug 21 '18 at 09:29
This is a very common FAQ. Any beginner-level C programming book will explain how to write proper macros, in the pre-processor chapter. It will dead certain mention this particular bug with operator precedence and lack of proper parenthesis. So go read that chapter. — Lundin, Aug 21 '18 at 09:30
@Lundin How about this one: [Is there a good reason for always enclosing a define in parentheses in C?](https://stackoverflow.com/questions/9081479/is-there-a-good-reason-for-always-enclosing-a-define-in-parentheses-in-c) — Andrew Henle, Aug 21 '18 at 09:52
@AndrewHenle Seems mostly focused on if parenthesis is still needed even though there's just one pre-processor token. I think we need something more basic to use as canonical dupe. — Lundin, Aug 21 '18 at 10:00
@Lundin Unfortunately, we killed *StackOverflow Documentation*... It would be useful in such cases — kyriakosSt, Aug 21 '18 at 10:56
@KyrSt No, one of many reasons why we killed it because it was partially filled with crap, written by confused beginners. For a high quality FAQ, there's the [C tag wiki](https://stackoverflow.com/tags/c/info), which is maintained and only points at high quality Q&A canonical dupes. — Lundin, Aug 21 '18 at 11:17

score 3 · Accepted Answer · answered Aug 21 '18 at 09:30

What happens is that

uint32_t num_rows = pager->file_length / ROW_SIZE;

is expanded into

uint32_t num_rows = pager->file_length / ID_SIZE + USERNAME_SIZE + EMAIL_SIZE;

so num_rows is evaluates to 0 + ID_SIZE + USERNAME_SIZE + EMAIL_SIZE, which is not a zero. Consider adding parentheses to your define:

#define ROW_SIZE (ID_SIZE + USERNAME_SIZE + EMAIL_SIZE)

score 2 · Answer 2 · answered Aug 21 '18 at 09:30

Macro expansion is token-based replacement. In other words,

pager->file_length / ROW_SIZE

expands to

pager->file_length / ID_SIZE + USERNAME_SIZE + EMAIL_SIZE

I believe you agree the above is not necessarily zero, especially if USERNAME_SIZE or EMAIL_SIZE are nonzero.

To solve this, put parentheses around the expression in the macro definition:

#define ROW_SIZE (ID_SIZE + USERNAME_SIZE + EMAIL_SIZE)

score 1 · Answer 3 · answered Aug 21 '18 at 10:10

1

It should be written like this:

#define ROW_SIZE (ID_SIZE + USERNAME_SIZE + EMAIL_SIZE)

answered Aug 21 '18 at 10:10

asdf

score 0 · Answer 4 · answered Aug 21 '18 at 09:30

You have to remember that preprocessor macros are basically query-replaced in the source code.

If you have e.g.

uint32_t num_rows = pager->file_length / ROW_SIZE;

what the compiler sees after preprocessing is

uint32_t num_rows = pager->file_length / ID_SIZE + USERNAME_SIZE + EMAIL_SIZE;

Due to the normal operator precedence that is equal to

uint32_t num_rows = (pager->file_length / ID_SIZE) + USERNAME_SIZE + EMAIL_SIZE;

which of course is not what you expect or want. Therefore it's important to always use extra parentheses in macros, as in

#define ROW_SIZE (ID_SIZE + USERNAME_SIZE + EMAIL_SIZE)

4 Answers4