Max value of an integer type?

[Frederick Gotham]

I'm writing a template, and I need to know the maximum value of a given
integer type at compile time. something like:

template<class NumType>
class Arb {
public:

static NumType const max = /* maximum value */;

};
If the template were only to be used with unsigned integer types, then
I'd do the following:

template<class NumType>
class Arb {
public:

static NumType const max = -1;

};
I need a compile-time constant which evaluates to the maximum value of an
integer type.

Any ideas?

If it weren't undefined behaviour to overflow a signed integer, I could
use a metaprogramming technique such as the following:

template<class T, bool overflow = false>
struct MaxIntVal {
private:

static T const internal = 1 + MaxIntVal<T,...

public:

static T const val = SCHAR_MAX + internal;
};
--

Frederick Gotham

[Alf P. Steinbach]

* Frederick Gotham:

I'm writing a template, and I need to know the maximum value of a given
integer type at compile time.

template< typename T struct Max;
template<struct Max<int>{ static int const value = INT_MAX; };
and so on

--
A: Because it messes up the order in which people normally read text.
Q: Why is it such a bad thing?
A: Top-posting.
Q: What is the most annoying thing on usenet and in e-mail?

[Ian Collins]

Frederick Gotham wrote:

>
I need a compile-time constant which evaluates to the maximum value of an
integer type.

Are you sure it has to be compile time? If not just use
std::numeric_limits<T>::max()
--
Ian Collins.

[Frederick Gotham]

I have a solution (although I intend on making it more compile-time-
efficient as I can assume that "digits" is at least 7).

#include<limits>

template< class T,
T shift_by = 0,
bool no_more_digits = shift_by == std::numeric_limits
<T>::digits

>

struct IntMax {
private:

static T const one = 1;

public:

static T const val = (one << shift_by) | IntMax<T,shift_by +
one>::val;

};

template<class T, T shift_by>
struct IntMax<T,shift_by,true{

static T const val = 0;
};
#include <iostream>
int main()
{
std::cout <<

"Max values\n"
"============\n\n"

" Signed char: " << (int)IntMax<signed char>::val <<

"\nSigned short: " << IntMax<short>::val <<

"\n Signed int: " << IntMax<int>::val <<

"\n Signed long: " << IntMax<long>::val;

}
--

Frederick Gotham

[Frederick Gotham]

Frederick Gotham posted:

>
I have a solution (although I intend on making it more compile-time-
efficient as I can assume that "digits" is at least 7).

Here's the more efficient version:
#include<limits>
template< class T,
T shift_by,
bool no_more_digits = shift_by == std::numeric_limits
<T>::digits

>

struct IntMax_Internal {
private:

static T const one = 1;

public:

static T const val = (one << shift_by) | IntMax_Internal<T,shift_by +
one>::val;

};

template<class T, T shift_by>
struct IntMax_Internal<T,shift_by,true{

static T const val = 0;
};

template< class T >
struct IntMax {
private:

static T const one_two_seven = 127;

public:

static T const val = one_two_seven | IntMax_Internal<T,7>::val;
};
#include <iostream>
int main()
{
std::cout <<

"Max values\n"
"==========\n\n"

" Unsigned char: " << (int)IntMax<unsigned char>::val <<

"\nUnsigned short: " << IntMax<unsigned short>::val <<

"\n Unsigned int: " << IntMax<unsigned>::val <<

"\n Unsigned long: " << IntMax<unsigned long>::val <<

"\n\n Signed char: " << (int)IntMax<signed char>::val <<

"\n Signed short: " << IntMax<short>::val <<

"\n Signed int: " << IntMax<int>::val <<

"\n Signed long: " << IntMax<long>::val;

}


--

Frederick Gotham

[gottlobfrege]

Frederick Gotham wrote:

template< class T,
T shift_by,
bool no_more_digits = shift_by == std::numeric_limits
<T>::digits

>

....

Although this would probably work 99% of the time, I think technically
that the standard does not require 2s compliment arithmetic (or
whatever it is called) - ie you can't assume setting all bits to 1
gives you the largest value.
As mentioned earlier, I think you would be better using INT_MAX, etc in
a bunch of specializations.
If you wanted to get really perverse, you might be able to do something
somehow similar to using bits - recursively double the value until it
doesn't get bigger, then (restore last largest value and) start adding
half as much, then 1/4, etc until you are not adding anything.

eg if max is 20 (strange max!)

search = 1;
try adding 2? =search = 3;
+ 4 ? =search = 7;
+ 8 ? =search = 15;
+ 16 ? =overflow!! - back off!
+ 8 ? =overflow!!
+ 4 ? =search = 19
+ 2 ? =overflow!
+ 1 ? =search = 20.
stop because you are at 1. If there was 'room' to add another 1, then
+ 2 would have worked.

Anyhow, perserve, and not really understandable by the next coder who
looks at it - even if that is you 2 years from now. But I suspect it
could be coded with templates.

Of course, I don't know what the standard says will happen when compile
time constants 'overflow'....

- Tony

[Harald van Dijk]

Frederick Gotham wrote:

I have a solution (although I intend on making it more compile-time-
efficient as I can assume that "digits" is at least 7).

#include<limits>

template< class T,
T shift_by = 0,
bool no_more_digits = shift_by == std::numeric_limits
<T>::digits

>

struct IntMax {
private:

static T const one = 1;

public:

static T const val = (one << shift_by) | IntMax<T,shift_by +
one>::val;

};

template<class T, T shift_by>
struct IntMax<T,shift_by,true{

static T const val = 0;
};

If you can use numeric_limits::digits, wouldn't this be simpler?

template <typename T>
struct intinfo {
static const T max = (static_cast<T>(1) <<
std::numeric_limits<T>::digits - 1) - 1 << 1 | 1;
};

Or are there some special cases this doesn't handle correctly?

[Rolf Magnus]

Ian Collins wrote:

Frederick Gotham wrote:

>>
I need a compile-time constant which evaluates to the maximum value of an
integer type.

Are you sure it has to be compile time? If not just use
std::numeric_limits<T>::max()

I guess there must be a good reason why this is not a compile-time constant,
but a function. I don't think that there could be an implementation where
those values change during runtime, so what is the reason to make it a
function?

[Frederick Gotham]

posted:

>
Frederick Gotham wrote:

>template< class T,
T shift_by,
bool no_more_digits = shift_by == std::numeric_limits
<T>::digits

> >

...

Although this would probably work 99% of the time, I think technically
that the standard does not require 2s compliment arithmetic (or
whatever it is called) - ie you can't assume setting all bits to 1
gives you the largest value.

"digits" gives you the amount of value bits, excluding the sign bit.

If the sign bit (if any) is zero, and if all the rest of the bits are 1,
then you have the max value.
--

Frederick Gotham

[Frederick Gotham]

=?utf-8?B?SGFyYWxkIHZhbiBExLNr?= posted:

If you can use numeric_limits::digits, wouldn't this be simpler?

template <typename T>
struct intinfo {
static const T max = (static_cast<T>(1) <<
std::numeric_limits<T>::digits - 1) - 1 << 1 | 1;
};

Or are there some special cases this doesn't handle correctly?

That's brilliant. Here's the updated code:

#include<limits>

template<class T>
struct IntMax {
private:

static T const MSBOnly = static_cast<T>(1) << std::numeric_limits
<T>::digits - 1;

public:

static T const val = MSBOnly | MSBOnly - 1;

};
#include <iostream>
int main()
{
std::cout <<

"Max values\n"
"==========\n\n"

" Unsigned char: " << (int)IntMax<unsigned char>::val <<

"\nUnsigned short: " << IntMax<unsigned short>::val <<

"\n Unsigned int: " << IntMax<unsigned>::val <<

"\n Unsigned long: " << IntMax<unsigned long>::val <<

"\n\n Signed char: " << (int)IntMax<signed char>::val <<

"\n Signed short: " << IntMax<short>::val <<

"\n Signed int: " << IntMax<int>::val <<

"\n Signed long: " << IntMax<long>::val;

}
--

Frederick Gotham

[Bo Persson]

"Frederick Gotham" <fg*******@SPAM.comskrev i meddelandet
news:F9*******************@news.indigo.ie...

posted:

>>
Frederick Gotham wrote:

>>template< class T,
T shift_by,
bool no_more_digits = shift_by == std::numeric_limits
<T>::digits
>

...

Although this would probably work 99% of the time, I think
technically
that the standard does not require 2s compliment arithmetic (or
whatever it is called) - ie you can't assume setting all bits to 1
gives you the largest value.

"digits" gives you the amount of value bits, excluding the sign bit.

If the sign bit (if any) is zero, and if all the rest of the bits
are 1,
then you have the max value.

This of course assumes that all the bits take part in the value
representation. Not a requirement.

You are now at 99.5% mark, I guess. :-)
There is a reason for the seemingly redundant values in
numeric_limits - implementations are allowed to have unusual values
for some of them.
Bo Persson

[Frederick Gotham]

Bo Persson posted:

This of course assumes that all the bits take part in the value
representation. Not a requirement.

18.2.1.2 numeric_limits members

static const int digits;

6 Number of radix digits that can be represented without change.
7 For built-in integer types, the number of non-sign bits in the
representation.
--

Frederick Gotham

[Rolf Magnus]

Bo Persson wrote:

>"digits" gives you the amount of value bits, excluding the sign bit.

If the sign bit (if any) is zero, and if all the rest of the bits
are 1,
then you have the max value.

This of course assumes that all the bits take part in the value
representation.

AFAICS, it doesn't. As Frederick wrote, "digits" gives you the amount of
_value bits_, not the total amount of bits.

[Alf P. Steinbach]

* Frederick Gotham:

=?utf-8?B?SGFyYWxkIHZhbiBExLNr?= posted:

>If you can use numeric_limits::digits, wouldn't this be simpler?

template <typename T>
struct intinfo {
static const T max = (static_cast<T>(1) <<
std::numeric_limits<T>::digits - 1) - 1 << 1 | 1;
};

Or are there some special cases this doesn't handle correctly?

That's brilliant. Here's the updated code:

#include<limits>

template<class T>
struct IntMax {
private:

static T const MSBOnly = static_cast<T>(1) << std::numeric_limits
<T>::digits - 1;

public:

static T const val = MSBOnly | MSBOnly - 1;

};

It's good fun to throw academic spanners into practical wheels, so,
where does the standard exclude a grey-code representation? ;-)

--
A: Because it messes up the order in which people normally read text.
Q: Why is it such a bad thing?
A: Top-posting.
Q: What is the most annoying thing on usenet and in e-mail?

[Frederick Gotham]

Alf P. Steinbach posted:

It's good fun to throw academic spanners into practical wheels, so,
where does the standard exclude a grey-code representation? ;-)

Only caveat that comes to mind is negative zero... but that shouldn't be an
issue here.

Unless you're talking about left-shifting a signed integer type... ?
(Not quite sure what you mean by "exclude a grey-code representation")
--

Frederick Gotham

[Alf P. Steinbach]

* Frederick Gotham:

Alf P. Steinbach posted:

>It's good fun to throw academic spanners into practical wheels, so,
where does the standard exclude a grey-code representation? ;-)

Only caveat that comes to mind is negative zero... but that shouldn't be an
issue here.

Unless you're talking about left-shifting a signed integer type... ?
(Not quite sure what you mean by "exclude a grey-code representation")

Frederick, I give you also (as I gave Protoman)... WIKIPEDIA!

<url: http://en.wikipedia.org/wiki/Grey_code>

Note the bit pattern for the highest value when using a given number of
bits.

Sorry for the typo.

Coming from a whaling country, I naturally prefer Earl Grey tea...

--
A: Because it messes up the order in which people normally read text.
Q: Why is it such a bad thing?
A: Top-posting.
Q: What is the most annoying thing on usenet and in e-mail?

[Frederick Gotham]

Alf P. Steinbach posted:

Note the bit pattern for the highest value when using a given number of
bits.

I was under the assumption that C++ had to store positive values as
follows:

0: 0000
1: 0001
2: 0010
3: 0011
4: 0100
5: 0101
6: 0110
7: 0111
8: 1000
9: 1001
10: 1010
11: 1011
12: 1100
13: 1101
14: 1110
15: 1111

Are you saying that:

(A) The machine could use another method such as "Gray code".
and
(B) My code could break on other systems, and thus isn't trully
portable?
--

Frederick Gotham

[Rolf Magnus]

Alf P. Steinbach wrote:

It's good fun to throw academic spanners into practical wheels, so,
where does the standard exclude a grey-code representation? ;-)

Hmm, I would have said 3.9.1/7, but after reading it again, it seems that
this actually doesn't prohibit a grey-code representation.

[Rolf Magnus]

Rolf Magnus wrote:

Alf P. Steinbach wrote:

>It's good fun to throw academic spanners into practical wheels, so,
where does the standard exclude a grey-code representation? ;-)

Hmm, I would have said 3.9.1/7, but after reading it again, it seems that
this actually doesn't prohibit a grey-code representation.

Ok, now I have one: 5.8/2.

[Frederick Gotham]

Rolf Magnus posted:

Ok, now I have one: 5.8/2.

The value of E1 << E2 is E1 (interpreted as a bit pattern) left-shifted E2
bit positions; vacated bits are zero-filled. If E1 has an unsigned type,
the value of the result is E1 multiplied by the quantity 2 raised to the
power E2, reduced modulo ULONG_MAX+1 if E1 has type unsigned long,
UINT_MAX+1 otherwise.
Is it the "vacated bits are zero-filled" part that you're talking about?

--

Frederick Gotham

[Alf P. Steinbach]

* Frederick Gotham:

Alf P. Steinbach posted:

>Note the bit pattern for the highest value when using a given number of
bits.

I was under the assumption that C++ had to store positive values as
follows:

0: 0000
1: 0001
2: 0010
3: 0011
4: 0100
5: 0101
6: 0110
7: 0111
8: 1000
9: 1001
10: 1010
11: 1011
12: 1100
13: 1101
14: 1110
15: 1111

Are you saying that:

(A) The machine could use another method such as "Gray code".
and
(B) My code could break on other systems, and thus isn't trully
portable?

It's just academic, but yes. §5.8/2 defines the result of << as a
bit-level operation, and as a value only for unsigned types. §5.8/3
defines the result of >as a bit-level operation and as a value both
for signed and unsigned types (the latter only for non-negative values).
The cases excluded from these definitions are, AFAICT, exactly those
where the definitions would otherwise be incompatible with grey-code...

--
A: Because it messes up the order in which people normally read text.
Q: Why is it such a bad thing?
A: Top-posting.
Q: What is the most annoying thing on usenet and in e-mail?

[Alf P. Steinbach]

* Rolf Magnus:

Rolf Magnus wrote:

>Alf P. Steinbach wrote:

>>It's good fun to throw academic spanners into practical wheels, so,
where does the standard exclude a grey-code representation? ;-)

Hmm, I would have said 3.9.1/7, but after reading it again, it seems that
this actually doesn't prohibit a grey-code representation.

Ok, now I have one: 5.8/2.

Nope, not as far as I can see: it's /very/ ingeniously worded so that
the cases that could be incompatible are simply not defined.

--
A: Because it messes up the order in which people normally read text.
Q: Why is it such a bad thing?
A: Top-posting.
Q: What is the most annoying thing on usenet and in e-mail?

[Alf P. Steinbach]

* Alf P. Steinbach:

grey-code

Sorry, again: _gray code_.

--
A: Because it messes up the order in which people normally read text.
Q: Why is it such a bad thing?
A: Top-posting.
Q: What is the most annoying thing on usenet and in e-mail?

[Rolf Magnus]

Frederick Gotham wrote:

Rolf Magnus posted:

>Ok, now I have one: 5.8/2.

The value of E1 << E2 is E1 (interpreted as a bit pattern) left-shifted E2
bit positions; vacated bits are zero-filled. If E1 has an unsigned type,
the value of the result is E1 multiplied by the quantity 2 raised to the
power E2, reduced modulo ULONG_MAX+1 if E1 has type unsigned long,
UINT_MAX+1 otherwise.
Is it the "vacated bits are zero-filled" part that you're talking about?

No, it's the "(interpreted as a bit pattern) left-shifted E2 bit positions"
in combination with the "multiplied by the quantity 2 raised to the power
E2". If your values are represented in gray code, a left-shift of the bit
pattern doesn't necessarily result in a mulitplication by 2.

[Alf P. Steinbach]

* Rolf Magnus:

Frederick Gotham wrote:

>Rolf Magnus posted:

>>Ok, now I have one: 5.8/2.

The value of E1 << E2 is E1 (interpreted as a bit pattern) left-shifted E2
bit positions; vacated bits are zero-filled. If E1 has an unsigned type,
the value of the result is E1 multiplied by the quantity 2 raised to the
power E2, reduced modulo ULONG_MAX+1 if E1 has type unsigned long,
UINT_MAX+1 otherwise.
Is it the "vacated bits are zero-filled" part that you're talking about?

No, it's the "(interpreted as a bit pattern) left-shifted E2 bit positions"
in combination with the "multiplied by the quantity 2 raised to the power
E2". If your values are represented in gray code, a left-shift of the bit
pattern doesn't necessarily result in a mulitplication by 2.

That's only for unsigned types. C++ unsigned integers are pure binary.
For signed integers there's much more leeway (formally).

--
A: Because it messes up the order in which people normally read text.
Q: Why is it such a bad thing?
A: Top-posting.
Q: What is the most annoying thing on usenet and in e-mail?

[Rolf Magnus]

Alf P. Steinbach wrote:

* Rolf Magnus:

>Rolf Magnus wrote:

>>Alf P. Steinbach wrote:

It's good fun to throw academic spanners into practical wheels, so,
where does the standard exclude a grey-code representation? ;-)
Hmm, I would have said 3.9.1/7, but after reading it again, it seems
that this actually doesn't prohibit a grey-code representation.

Ok, now I have one: 5.8/2.

Nope, not as far as I can see: it's /very/ ingeniously worded so that
the cases that could be incompatible are simply not defined.

What's not defined? It says clearly that the bit pattern is shifted left by
the specified number of bits and that this will (for unsigned types) result
in a multiplication by 2 raised to the value of the rhs. That's not
possible with Gray code, so unsigned types cannot use Gray code. Since a
bit pattern for a signed value must be the same as for the corresponding
unsigned value, signed types neither can use Gray code.

[Rolf Magnus]

Alf P. Steinbach wrote:

* Rolf Magnus:

>Frederick Gotham wrote:

>>Rolf Magnus posted:
Ok, now I have one: 5.8/2.
The value of E1 << E2 is E1 (interpreted as a bit pattern) left-shifted
E2 bit positions; vacated bits are zero-filled. If E1 has an unsigned
type, the value of the result is E1 multiplied by the quantity 2 raised
to the power E2, reduced modulo ULONG_MAX+1 if E1 has type unsigned
long, UINT_MAX+1 otherwise.
Is it the "vacated bits are zero-filled" part that you're talking about?

No, it's the "(interpreted as a bit pattern) left-shifted E2 bit
positions" in combination with the "multiplied by the quantity 2 raised
to the power E2". If your values are represented in gray code, a
left-shift of the bit pattern doesn't necessarily result in a
mulitplication by 2.

That's only for unsigned types. C++ unsigned integers are pure binary.
For signed integers there's much more leeway (formally).

Only for the negative values. So I guess a signed integer could use Gray
code for the negative part. That would be extremely odd, but seems to be
allowed by the standard.

[Alf P. Steinbach]

* Rolf Magnus:

Alf P. Steinbach wrote:

>* Rolf Magnus:

>>Frederick Gotham wrote:

Rolf Magnus posted:
Ok, now I have one: 5.8/2.
The value of E1 << E2 is E1 (interpreted as a bit pattern) left-shifted
E2 bit positions; vacated bits are zero-filled. If E1 has an unsigned
type, the value of the result is E1 multiplied by the quantity 2 raised
to the power E2, reduced modulo ULONG_MAX+1 if E1 has type unsigned
long, UINT_MAX+1 otherwise.
Is it the "vacated bits are zero-filled" part that you're talking about?
No, it's the "(interpreted as a bit pattern) left-shifted E2 bit
positions" in combination with the "multiplied by the quantity 2 raised
to the power E2". If your values are represented in gray code, a
left-shift of the bit pattern doesn't necessarily result in a
mulitplication by 2.

That's only for unsigned types. C++ unsigned integers are pure binary.
For signed integers there's much more leeway (formally).

Only for the negative values. So I guess a signed integer could use Gray
code for the negative part. That would be extremely odd, but seems to be
allowed by the standard.

No, it's for any (signed) value. Left shift doesn't define the numeric
value. And right shift >does define the value but is all OK.

--
A: Because it messes up the order in which people normally read text.
Q: Why is it such a bad thing?
A: Top-posting.
Q: What is the most annoying thing on usenet and in e-mail?

[Alf P. Steinbach]

* Rolf Magnus:

Alf P. Steinbach wrote:

>* Rolf Magnus:

>>Rolf Magnus wrote:

Alf P. Steinbach wrote:

It's good fun to throw academic spanners into practical wheels, so,
where does the standard exclude a grey-code representation? ;-)
Hmm, I would have said 3.9.1/7, but after reading it again, it seems
that this actually doesn't prohibit a grey-code representation.
Ok, now I have one: 5.8/2.

Nope, not as far as I can see: it's /very/ ingeniously worded so that
the cases that could be incompatible are simply not defined.

What's not defined? It says clearly that the bit pattern is shifted left by
the specified number of bits and that this will (for unsigned types) result
in a multiplication by 2 raised to the value of the rhs. That's not
possible with Gray code, so unsigned types cannot use Gray code. Since a
bit pattern for a signed value must be the same as for the corresponding
unsigned value, signed types neither can use Gray code.

No, the bit pattern for a signed value must not necessarily be the same
as for the corresponding unsigned value. In particular, that very
obviously doesn't hold for negative values. And by extension, it
doesn't hold for non-negative values either (if you're looking at the
same paragraph which has been discussed to death in this connection
before [the solution, if it may be called that, is then to look at the
definitions of terms] ;-)).

--
A: Because it messes up the order in which people normally read text.
Q: Why is it such a bad thing?
A: Top-posting.
Q: What is the most annoying thing on usenet and in e-mail?

[Rolf Magnus]

Alf P. Steinbach wrote:

* Rolf Magnus:

>Alf P. Steinbach wrote:

>>* Rolf Magnus:
Rolf Magnus wrote:

Alf P. Steinbach wrote:
>
>It's good fun to throw academic spanners into practical wheels, so,
>where does the standard exclude a grey-code representation? ;-)
Hmm, I would have said 3.9.1/7, but after reading it again, it seems
that this actually doesn't prohibit a grey-code representation.
Ok, now I have one: 5.8/2.
Nope, not as far as I can see: it's /very/ ingeniously worded so that
the cases that could be incompatible are simply not defined.

What's not defined? It says clearly that the bit pattern is shifted left
by the specified number of bits and that this will (for unsigned types)
result in a multiplication by 2 raised to the value of the rhs. That's
not possible with Gray code, so unsigned types cannot use Gray code.
Since a bit pattern for a signed value must be the same as for the
corresponding unsigned value, signed types neither can use Gray code.

No, the bit pattern for a signed value must not necessarily be the same
as for the corresponding unsigned value. In particular, that very
obviously doesn't hold for negative values. And by extension, it
doesn't hold for non-negative values either (if you're looking at the
same paragraph which has been discussed to death in this connection
before [the solution, if it may be called that, is then to look at the
definitions of terms] ;-)).

You must be talking about:

"The range of nonnegative values of a signed integer type is a subrange of
the corresponding unsigned integer type, and the value representation of
each corresponding signed/unsigned type shall be the same."

Not sure what's wrong about that, though. Doesn't it mean that the same
value in a corresponding signed and unsigned type must have the same bit
pattern? Obviously, that's only possible for the subrange that is covered
by both.

[Alf P. Steinbach]

* Rolf Magnus -Alf P. Steinbach:

>
You must be talking about:

"The range of nonnegative values of a signed integer type is a subrange of
the corresponding unsigned integer type, and the value representation of
each corresponding signed/unsigned type shall be the same."

Yep.

Not sure what's wrong about that, though. Doesn't it mean that the same
value in a corresponding signed and unsigned type must have the same bit
pattern?

Nope, partly because:

Obviously, that's only possible for the subrange that is covered
by both.

and partly because of earlier definition of "value representation".

--
A: Because it messes up the order in which people normally read text.
Q: Why is it such a bad thing?
A: Top-posting.
Q: What is the most annoying thing on usenet and in e-mail?

[Bo Persson]

"Rolf Magnus" <ra******@t-online.deskrev i meddelandet
news:e8*************@news.t-online.com...

Bo Persson wrote:

>>"digits" gives you the amount of value bits, excluding the sign
bit.

If the sign bit (if any) is zero, and if all the rest of the bits
are 1,
then you have the max value.

This of course assumes that all the bits take part in the value
representation.

AFAICS, it doesn't. As Frederick wrote, "digits" gives you the
amount of
_value bits_, not the total amount of bits.

Does too! :-)

How do we know exactly what bits of the object takes part in the value
representation? Is the sign bit always the first bit? Where are the
pad bits?

If we always assume 32 bits, two's complement, and no pad bits, we can
skip all this. If we want to be totally portable, we cannot assume any
of it.
Part of the reason for me being so picky, is that on my CS courses we
used a Univac 1100 - 36 bit, one's complement. Compatible machines
still exist!

Bo Persson

[Frederick Gotham]

Bo Persson posted:

Does too! :-)

How do we know exactly what bits of the object takes part in the value
representation? Is the sign bit always the first bit? Where are the
pad bits?

That's the compiler's job to worry about. We just give it the value of the
bit we want, rather than the bit index, and it works out the rest.

Furthermore, the Standard guarantees that:

x << 1

is equal to:

x * 2
--

Frederick Gotham

[Michiel.Salters]

Alf P. Steinbach wrote:
the bit pattern for a signed value must not necessarily be the same

as for the corresponding unsigned value. In particular, that very
obviously doesn't hold for negative values. And by extension, it
doesn't hold for non-negative values either

I don't agree with your extension. Take a simple 16 bit int. The
common value subrange is [0,32767], for which bits 0-15 are defined
by the unsigned int representation. For this range, and this range
only,
the representation of int and unsigned int must match. The fact that
the ranges [-32768,-1] for int and [32768,65536] for unsigned are not
common means the bit patterns for these ranges may overlap
(must, if only 16 bits are available) but that only says something
about
bit patterns where bit 15 is set.

[Frederick Gotham]

I have a better solution.

The highest value is equal to:
(radix ^ digits) - 1
(I thought it might work for floating point types aswell, but it
doesn't.)

First I start off with a metaprogramming template to raise a number to a
positive integer power, and work from there:

#include <limits>

template<class T, unsigned long base, unsigned long exp,
bool exp_non_zero = !!exp>
struct IntPow {

static T const val = base * IntPow<T,base,exp-1>::val;
};

template<class T, unsigned long base, unsigned long exp>
struct IntPow<T,base,exp,false{

static T const val = 1;
};

template<class T>
struct IntMax {

typedef std::numeric_limits<TLim;

static T const val =
IntPow<T, Lim::radix, Lim::digits>::val - 1;

};

#include <iostream>
int main()
{
std::cout <<

"Max values\n"
"==========\n\n"

" unsigned char: " << (int)IntMax<unsigned char>::val <<

"\nunsigned short: " << IntMax<unsigned short>::val <<

"\n unsigned int: " << IntMax<unsigned>::val <<

"\n unsigned long: " << IntMax<unsigned long>::val <<

"\n\n signed char: " << (int)IntMax<signed char>::val <<

"\n signed short: " << IntMax<short>::val <<

"\n signed int: " << IntMax<int>::val <<

"\n signed long: " << IntMax<long>::val;

/*
std::cout <<

"\n float: " << IntMax<float>::val <<

"\n double: " << IntMax<float>::val <<

"\n long double: " << IntMax<float>::val;
*/

}
--

Frederick Gotham

[Alf P. Steinbach]

* Mi*************@tomtom.com:

Alf P. Steinbach wrote:
the bit pattern for a signed value must not necessarily be the same

>as for the corresponding unsigned value. In particular, that very
obviously doesn't hold for negative values. And by extension, it
doesn't hold for non-negative values either

I don't agree with your extension. Take a simple 16 bit int. The
common value subrange is [0,32767], for which bits 0-15 are defined
by the unsigned int representation. For this range, and this range
only,
the representation of int and unsigned int must match.

What is the argument?

--
A: Because it messes up the order in which people normally read text.
Q: Why is it such a bad thing?
A: Top-posting.
Q: What is the most annoying thing on usenet and in e-mail?