Missing std::sign

[Matthew Fioravante]

The math library is missing the signum function.

//Returns 1 if x > 0, 0 if x == 0, or -1 if x < 0
template <typename T> constexpr T sign(T x);

Here are a whole bunch of people arguing about ways to implement it. Be nice if there was a standard version with optimal efficiency on each platform.
http://stackoverflow.com/questions/1903954/is-there-a-standard-sign-function-signum-sgn-in-c-c

It is possible to add new functions to the <cmath> header without getting them in the C standard first?

[Howard Hinnant]

Yes, as long as the spec clearly states that the signature is not guaranteed to be available from <math.h>. Unofficially, but from a practical standpoint, we own <cmath>, and the C committee owns <math.h>. Whenever either of the two committees tries to specify the other’s header (and both committees have attempt to do so in the past), a little bit more chaos escapes into the universe, and is nearly impossible to recapture. To this day we suffer from such attempts in the past.

Howard

[Gabriel Dos Reis]

Howard Hinnant <howard....@gmail.com> writes:

| guaranteed to be available from <math.h>. Unofficially, but from a
| practical standpoint, we own <cmath>, and the C committee owns
| <math.h>. Whenever either of the two committees tries to specify the
| other's header (and both committees have attempt to do so in the
| past), a little bit more chaos escapes into the universe, and is
| nearly impossible to recapture. To this day we suffer from such
| attempts in the past.

Amen to that.

Furthermore, LWG should stop Pavlovian reflexive adoption of everything
WG14 invents as <cxxx>. Case in point: <cstdalign>.

-- Gaby

[Matthew Fioravante]

Maybe it would make sense to have a C++ <math> header which includes <cmath> and also has C++ specific math functions? Similarly for other headers. The header <cmath> sounds like a C compatibility header, not the primary header for the C++ math library.

[Howard Hinnant]

On Aug 27, 2014, at 8:13 PM, Matthew Fioravante <fmatth...@gmail.com> wrote:

> Maybe it would make sense to have a C++ <math> header which includes <cmath> and also has C++ specific math functions? Similarly for other headers. The header <cmath> sounds like a C compatibility header, not the primary header for the C++ math library.

No, just put the C++ specific math functions in <cmath>, and then explicitly state that they do not appear in <math.h> (unless the C committee says otherwise).

We don’t need more headers. And there is no problem in the C++ committee adding functionality to <cmath>. We only get in trouble when the C++ committee adds (or subtracts) functionality to <math.h>, or when the C committee adds (or subtracts) functionality to <cmath>.

Howard

[Gabriel Dos Reis]

Howard Hinnant <howard....@gmail.com> writes:

| On Aug 27, 2014, at 8:13 PM, Matthew Fioravante <fmatth...@gmail.com> wrote:
|
| > Maybe it would make sense to have a C++ <math> header which includes
| > <cmath> and also has C++ specific math functions? Similarly for
| > other headers. The header <cmath> sounds like a C compatibility
| > header, not the primary header for the C++ math library.
|
| No, just put the C++ specific math functions in <cmath>, and then
| explicitly state that they do not appear in <math.h> (unless the C
| committee says otherwise).

Agreed. We already have too many C++ headers that do "math" stuff and I
always forget which one to include -- <cmath>, <cstdlib>, <numerics>,
<algorithm>, <complex>, <valarray>, etc.

| We don’t need more headers. And there is no problem in the C++
| committee adding functionality to <cmath>. We only get in trouble
| when the C++ committee adds (or subtracts) functionality to <math.h>,
| or when the C committee adds (or subtracts) functionality to <cmath>.
|
| Howard

-- Gaby