This is one of the last dmd changes: http://www.dsource.org/projects/dmd/changeset/557

cent/ucent are probably easy to implement, their implementation can require little code on 64 bit systems, and operations done among them are probably fast on 64 bit systems.

128 bit unsigned numbers can represent:
340_282_366_920_938_463_463_374_607_431_768_211_455
Many normal programs don't overflow such numbers. A cent number can also be used to store safely the result of many operations done on 64 bit values. And generally I like the idea of having them...

But while I know of few scientific computations situations where quadruple precision floating point numbers can be useful, in practice I don't know of situations where cent/ucent can be useful.

One possible situation where ucent can be useful is to perform bitwise operations on 128-bits-wide SSE registers. But a good D compiler can use exactly the same CPU instructions when those operators are used among uint[4] fixed-sized arrays.

So unless some significant application can be found, I suggest to not add cent/ucent. All features in the language need a purpose, adding useless things is bad. They can be kept among reserved keywords (as 'macro') for possible 128-bit CPUs.

On the other hand instead of cent/ucent I have several situations (not related with cryptography) where I'd like in D a built-in "bint" or "bigint", that is the multi-precision integers (currently in a std.bigint module... well, the module was present in 2.046, I can't see it in Phobos of 2.047, I don't know why). Having it built-in allows for better multi-precision literals.

(There are other number types that probably are more useful than cent/ucent, like a decimal floating point (IEEE standard 854-1987), useful for concurrency.)

Bye,
bearophile

bearophile wrote:
> This is one of the last dmd changes:
> http://www.dsource.org/projects/dmd/changeset/557
> 
> cent/ucent are probably easy to implement, their implementation can require little code on 64 bit systems, and operations done among them are probably fast on 64 bit systems.
<snip>

Whether it's implemented any time soon or not, I'd like the "feature" whereby trying to use cent/ucent is a _syntax_ error to be finally removed.  In the current state of things, making code forward compatible is a nightmare, and the later it's left the less backward compatible any code using cent/ucent can expect to be.

http://d.puremagic.com/issues/show_bug.cgi?id=785

Unfortunately it's now too late for libraries targeting D 1.0 to make use of it.  But still, would it make sense to drop the fix into D1, given that there is still work to do on the spec anyway?

Stewart.

Stewart Gordon:
> Unfortunately it's now too late for libraries targeting D 1.0 to make use of it.  But still, would it make sense to drop the fix into D1, given that there is still work to do on the spec anyway?

D1 is mostly in debug mode now. In my opinion specs and other more complex jobs about D1 are going to be delayed indefinitely by Walter. So I think future changes to D1 will need to come from the community, like Tango users interested in keeping D1 alive.

Bye,
bearophile

bearophile wrote:

> This is one of the last dmd changes: http://www.dsource.org/projects/dmd/changeset/557
> 
> cent/ucent are probably easy to implement, their implementation can require little code on 64 bit systems, and operations done among them are probably fast on 64 bit systems.
> 
> 128 bit unsigned numbers can represent:
> 340_282_366_920_938_463_463_374_607_431_768_211_455
> Many normal programs don't overflow such numbers. A cent number can also
> be used to store safely the result of many operations done on 64 bit
> values. And generally I like the idea of having them...
> 
> But while I know of few scientific computations situations where quadruple precision floating point numbers can be useful, in practice I don't know of situations where cent/ucent can be useful.
> 
> One possible situation where ucent can be useful is to perform bitwise operations on 128-bits-wide SSE registers. But a good D compiler can use exactly the same CPU instructions when those operators are used among uint[4] fixed-sized arrays.
> 
> So unless some significant application can be found, I suggest to not add cent/ucent. All features in the language need a purpose, adding useless things is bad. They can be kept among reserved keywords (as 'macro') for possible 128-bit CPUs.
> 
> On the other hand instead of cent/ucent I have several situations (not related with cryptography) where I'd like in D a built-in "bint" or "bigint", that is the multi-precision integers (currently in a std.bigint module... well, the module was present in 2.046, I can't see it in Phobos of 2.047, I don't know why). Having it built-in allows for better multi-precision literals.
> 
> (There are other number types that probably are more useful than
> cent/ucent, like a decimal floating point (IEEE standard 854-1987), useful
> for concurrency.)
> 
> Bye,
> bearophile

cent and ucent are not currently implemented and have no need to be implemented at this point. They are reserved for when they will be deemed useful enough to implement (perhaps when we actually have 128-bit machines). In the meantime, I see no need to worry about it. They'll get implemented when it's appropriate. For now, we can just leave them as reserved and not worry about them. The fact that they exist makes it possible to implement them later without breaking backwards compatibility.

- Jonathan M Davis

Jonathan M Davis:
> cent and ucent are not currently implemented and

Isn't changeset 557 showing some work to implement them?

Bye,
bearophile

bearophile wrote:
> Jonathan M Davis:
>> cent and ucent are not currently implemented and
> 
> Isn't changeset 557 showing some work to implement them?
> 
> Bye,
> bearophile

No, it's back-end only. The opcodes emitted in 32-bit code for long and ulong are exactly the same as the ones for cent and ucent in 64-bit code. Including them is *much* simpler than not including them.

Don:
> Including them is *much* simpler than not including them.

Thank you Don.

Bye,
bearophile

bearophile wrote:
> Stewart Gordon:
>> Unfortunately it's now too late for libraries targeting D 1.0 to make use of it.  But still, would it make sense to drop the fix into D1, given that there is still work to do on the spec anyway?
> 
> D1 is mostly in debug mode now. In my opinion specs and other more
> complex jobs about D1 are going to be delayed indefinitely by Walter.
> So I think future changes to D1 will need to come from the community,
> like Tango users interested in keeping D1 alive.

If that's the case, it's time Walter's pretences of interest in making D implementable by third parties be removed from the docs.

Stewart.

bearophile wrote:
> Stewart Gordon:
>> Unfortunately it's now too late for libraries targeting D 1.0 to make use of it.  But still, would it make sense to drop the fix into D1, given that there is still work to do on the spec anyway?
> 
> D1 is mostly in debug mode now. In my opinion specs and other more complex jobs about D1 are going to be delayed indefinitely by Walter. So I think future changes to D1 will need to come from the community, like Tango users interested in keeping D1 alive.
> 
> Bye,
> bearophile

Sounds to me in other words that

Though D1 is now brain dead (c.f. more complex jobs about D1 are going to be delayed indefinitely by Walter),
D1 might be kept on life support (i.e. via an intravenous drip of Tango juice or similar)

Either way it is pretty much apparent that D1 has already died of natural causes.

Cheers
Justin Johansson

Justin Johansson wrote:
<snip>
> Though D1 is now brain dead (c.f. more complex jobs about D1 are going to be delayed indefinitely by Walter),
<snip>

What "more complex jobs" would they be, let alone ones that don't equally affect D2?

Stewart.