Template version of std::abs

Question

Here lists the current overloads of std::abs in C++. I'm wondering why not just define the following template and let go all the ugly C-style overloads?

template <typename T> inline
T abs(const T& v) { return v < 0 ? -v : v; }

@MariusBancila Then it won't compile for non-compatible types. — Neil Kirk, Apr 02 '15 at 11:15
I also would like to know the answer to the question. As a template it would allow abs to work with your own classes that support the appropriate operators. — Neil Kirk, Apr 02 '15 at 11:16
It would make it a lot easier to write `&abs` - currently that's a pain. — MSalters, Apr 02 '15 at 11:25
@MariusBancila: Comparing a string to an integer would not compile but would admittedly give an unintellegible error message. However, you can add something like `std::enable_if, void>` and you'll get an error like "Candidate template ignored because..." which is very descriptive. — Damon, Apr 02 '15 at 11:43
Also see [Is std::abs(0u) ill-formed?](http://stackoverflow.com/q/29750946/1708801) — Shafik Yaghmour, Apr 22 '15 at 19:21

score 22 · Accepted Answer · edited Jan 12 '22 at 15:55

22

See LWG issue 2192. Currently, std::abs(x-y) < 2 fails if x and y are unsigned. This catches a subtle programming error. With the proposed change, it compiles but does entirely the wrong thing. abs(3u-4u) would be much larger than 2, in fact it's UINT_MAX.

edited Jan 12 '22 at 15:55

Patrik

2,695
1
21
36

answered Apr 02 '15 at 11:41

MSalters

173,980
10
155
350

@MohitJain `0u - 1u == UINT_MAX`, and `UINT_MAX == 2^(whatever) - 1`. – Quentin Apr 02 '15 at 12:00
2

That is a great example and issue 2192 is an interesting read. – Shafik Yaghmour Apr 02 '15 at 12:43
Well, `std::abs(3u - 4u)` compiles. See http://www.cplusplus.com/reference/cmath/abs/. – Lingxi Apr 02 '15 at 13:28
Still, I do not get it why the overloads for integral types return a `double` instead of the original integral type. – Lingxi Apr 02 '15 at 13:33
@Lingxi: Wrong page? http://www.cplusplus.com/reference/cstdlib/abs/ . Or look at a better site: http://en.cppreference.com/w/cpp/numeric/math/abs – MSalters Apr 02 '15 at 13:37
It's counter-intuitive to me that the overloads for signed integral types are put in `` other than ``. I just thought that all math functions are put int the latter header. – Lingxi Apr 02 '15 at 13:58
@Lingxi: That's another issue, [LWG 2294](http://www.open-std.org/jtc1/sc22/wg21/docs/lwg-active.html#2294) – MSalters Apr 02 '15 at 14:03

score 11 · Answer 2 · answered Apr 02 '15 at 11:20

11

This suffers the usual problem of matching everything.

An example of a type for which abs makes sense but this implementation does not is complex<double>.

answered Apr 02 '15 at 11:20

3

There is an `std::abs` overload for complex numbers that would be found by overload resolution and preferred, since it is more specialized (in terms of partial ordering) than the above. – Columbo Apr 02 '15 at 11:21
@Columbo: it does answer the question why you can't eliminate _all_ current overloads. – MSalters Apr 02 '15 at 11:23
@MSalters Well you could specialise the template for Complex. – Neil Kirk Apr 02 '15 at 11:25
3

@Columbo: The point isn't that `complex` won't work as is: the point is to demonstrate the definition in the OP can be flat out wrong for reasonable types. Similar user-defined types would fail to compile (or worse, silently compile a *wrong* implementation). – Apr 02 '15 at 11:28
@Hurkyl I didn't try to criticize your answer anyway, that was more of a side-note - this template and overloads for e.g. `complex` go together quite well :) – Columbo Apr 02 '15 at 11:29
It gets to a point where you can't protect the user from calling functions they don't understand. – Neil Kirk Apr 02 '15 at 11:31
@NeilKirk:That specialization would be a partial specialization (`std::complex` itself is a template) and you cannot partially specialize functions - precisely because you can have overloads. – MSalters Apr 02 '15 at 11:34
@Hurkyl What you said also applies to `std::swap`. It do not think it is a problem. – Lingxi Apr 02 '15 at 11:34
@Hurkyl But you do answer why we should live with the overloads :) – Lingxi Apr 02 '15 at 11:38
This could be solved [with a little bit of ADL](http://codereview.stackexchange.com/q/60140/15094) injected into the function. – Morwenn Apr 02 '15 at 14:07
But aren't these problems similar for std::less? If you feed it with rubbish you get rubbish. – gast128 Mar 20 '19 at 10:21

score 1 · Answer 3 · answered Apr 02 '15 at 11:14

1

Because type 'T' can access any data type including char. So what are you expecting if someone will pass a char to the abs function. :)

answered Apr 02 '15 at 11:14

Nihar

347
2
16

9

The same as for int, except it will only be 8-bits. – Neil Kirk Apr 02 '15 at 11:15
yes @NeilKirk you are correct as c++ will do a ascii value manipulation we will get a result . But that is not a output what an end user want. – Nihar Apr 02 '15 at 11:17
3

No, C++ won't do an " ASCII" manipulation. it will without a doubt do a numerical manipulation. The biggest issue in fact would be that `char` may be unsigned, which would give rather unexpected results (!) – MSalters Apr 02 '15 at 11:19
It's common to use char types to hold 8-bit data as, well, there isn't any alternative, at least not in the past. Performing abs on `char` would be of questionable usefulness however, as its signness is implementation defined. – Neil Kirk Apr 02 '15 at 11:20

Template version of std::abs

3 Answers3

Linked