This is another small imperfection we should get rid of.
The floating point types have a property called ".min", but unlike the integer ".min", it's not the minimum!
This naming stupidity is inherited from C++. The minimum float is -float.max. Instead, float.min is the minimum representable positive normalized number.
(BTW there are also representable "subnormal" numbers between 0 and float.min. So the name 'min' is _completely_ inappropriate, it's not even the minimum absolute value).

This misnaming is bad because (a) it causes confusion; and (b) it interfere with generic code, requiring special cases.
We should rename this while we have the chance. I don't think we should depart too far from the C/C++ name, but anything other than ".min" will work. I propose:

real.min ----> real.min_normal

Comments:
(1) Embedded underscores are probably not ideal, but they are already in use in the floating point properties mant_dig, and the rarely used max_10_exp, max_exp, min_10_exp, min_exp. They're also used in foreach_reverse, and in the built-in version identifiers, such as X86_64 and D_InlineAsm_X86. So there seems no reason to avoid them here. But if you have a much better idea for a name, speak now!
(2) There's no need for float.max_normal. float.max truly is the maximum representable number (excluding infinity).
Although float.min sounds like some kind of inverse of float.max, it is not. (The relationship is: X.max * X.min = 4.0 for binary types, 10.0 for decimal types).
Changing float.min -> float.min_normal will remove that illusion.

If there is no objection to this, I will create a patch. It's very simple.

I agree, when i had to type float.min i was confused it it was the smallest subnormal or smallest normal float

Don wrote:
> This is another small imperfection we should get rid of.
> The floating point types have a property called ".min", but unlike the integer ".min", it's not the minimum!
> This naming stupidity is inherited from C++. The minimum float is -float.max. Instead, float.min is the minimum representable positive normalized number.
> (BTW there are also representable "subnormal" numbers between 0 and float.min. So the name 'min' is _completely_ inappropriate, it's not even the minimum absolute value).
> 
> This misnaming is bad because (a) it causes confusion; and (b) it interfere with generic code, requiring special cases.
> We should rename this while we have the chance. I don't think we should depart too far from the C/C++ name, but anything other than ".min" will work. I propose:
> 
> real.min ----> real.min_normal
> 
> Comments:
> (1) Embedded underscores are probably not ideal, but they are already in use in the floating point properties mant_dig, and the rarely used max_10_exp, max_exp, min_10_exp, min_exp. They're also used in foreach_reverse, and in the built-in version identifiers, such as X86_64 and D_InlineAsm_X86. So there seems no reason to avoid them here. But if you have a much better idea for a name, speak now!


After I read your floating-point article on the DMD web site that name has been bothering me too. I vote for changing it.

I wouldn't worry about the underscore, as there is already precedence for using them in keywords and built-in properties. If people don't like that, we should also change the others.

Would it make sense to redefine real.min to mean -real.max? It would at least be consistent with the meaning of int.min, and perhaps it would find use in generic code.

-Lars

Don Wrote:

> The floating point types have a property called ".min", but unlike the integer ".min", it's not the minimum!

That bothered me too when I was writing about it before; I felt embarrassed when explaining that it was not actually the minimum. :)

Ali

Don:

> such as X86_64 and D_InlineAsm_X86<

I don't like the identifier that denotes D2 code ("D_Version2").


> If there is no objection to this, I will create a patch. It's very simple.

I usually like all your proposals about floating point numbers :-)

Bye,
bearophile

Don wrote:
> This is another small imperfection we should get rid of.
> The floating point types have a property called ".min", but unlike the integer ".min", it's not the minimum!

I've always hated that with a passion. Yes Don, great initiative. Let's change that crap once and for all. I'd consider using minNormal, but you make a good point that _ is already present.

++vote for patching. There are a couple of places in Phobos that need changing, I can operate them.


Andrei

bearophile wrote:
> Don:
> 
>> such as X86_64 and D_InlineAsm_X86<
> 
> I don't like the identifier that denotes D2 code ("D_Version2").

I never use it either, this identifier is rather pointless. Both compilers will still parse both conditions and errors on the branch for the other version.

>> If there is no objection to this, I will create a patch. It's very simple.
> 
> I usually like all your proposals about floating point numbers :-)

I agree, its a good change! I would also change float.min to be an alias of -float.max.

Jeremie

Jeremie Pelletier:

> I never use it either, this identifier is rather pointless. Both compilers will still parse both conditions and errors on the branch for the other version.

Sorry, as usual I want not precise enough, I meant I don't like the name of that identifier, but I use that functionality, I use version(D_Version2) when I want to write code that works with both Phobos1 and Phobos2 (so for example I import to!(int, string) from Phobos2 and toInt from Phobos1).

Bye,
bearophile

Jeremie Pelletier wrote:
> I agree, its a good change! I would also change float.min to be an alias of -float.max.

Unfortunately, we can't do that yet, it would silently change the meaning of existing code. Hopefully it can be done eventually.

Don wrote:
> This is another small imperfection we should get rid of.
> The floating point types have a property called ".min", but unlike the integer ".min", it's not the minimum!

> This misnaming is bad because (a) it causes confusion; and (b) it interfere with generic code, requiring special cases.
> We should rename this while we have the chance. I don't think we should depart too far from the C/C++ name, but anything other than ".min" will work. I propose:
> 
> real.min ----> real.min_normal
> If there is no objection to this, I will create a patch. It's very simple.

Patch is in bugzilla 3446. It took about 2 minutes to do.