dchar v11=dchar.max;
	auto vp11 = [v11];
	
	auto v2 = cast(ubyte[]) (vp11);   //v2.length=4
	auto v22 = cast(ubyte[])( [v11]); //v2.length=1


y22.length <> v2.length

On 3/18/18 4:34 AM, sdvcn wrote:
> dchar v11=dchar.max;
>      auto vp11 = [v11];
> 
>      auto v2 = cast(ubyte[]) (vp11);   //v2.length=4
>      auto v22 = cast(ubyte[])( [v11]); //v2.length=1


This seems like a bug to me.

It appears that v22 has truncated v11 to a byte and made only a single byte array out of it.

-Steve

On Sunday, March 18, 2018 13:10:28 Steven Schveighoffer via Digitalmars-d wrote:
> On 3/18/18 4:34 AM, sdvcn wrote:
> > dchar v11=dchar.max;
> >
> >      auto vp11 = [v11];
> >
> >      auto v2 = cast(ubyte[]) (vp11);   //v2.length=4
> >      auto v22 = cast(ubyte[])( [v11]); //v2.length=1
>
> This seems like a bug to me.
>
> It appears that v22 has truncated v11 to a byte and made only a single byte array out of it.

Except that that's precisely how you usually get an array any integral type smaller than an integer. e.g.

auto arr = cast(ubyte[])([1, 2, 3, 4]);

In this case, you could do

ubyte[] arr = [1, 2, 3, 4];

instead, but if you're not dealing with an initializaton or assignment like this (e.g. you're passing the array to a functon), then the cast is the way you do it. Normally, you do it with integer literals, and I could see an argument that it shouldn't allow it without VRP being used to make it work, but it _is_ a cast, and casts are a bit of a blunt instrument.

So, I really don't think that it's a bug.

- Jonathan M Davis

On 3/18/18 2:24 PM, Jonathan M Davis wrote:
> On Sunday, March 18, 2018 13:10:28 Steven Schveighoffer via Digitalmars-d
> wrote:
>> On 3/18/18 4:34 AM, sdvcn wrote:
>>> dchar v11=dchar.max;
>>>
>>>       auto vp11 = [v11];
>>>
>>>       auto v2 = cast(ubyte[]) (vp11);   //v2.length=4
>>>       auto v22 = cast(ubyte[])( [v11]); //v2.length=1
>>
>> This seems like a bug to me.
>>
>> It appears that v22 has truncated v11 to a byte and made only a single
>> byte array out of it.
> 
> Except that that's precisely how you usually get an array any integral type
> smaller than an integer. e.g.
> 
> auto arr = cast(ubyte[])([1, 2, 3, 4]);
> 
> In this case, you could do
> 
> ubyte[] arr = [1, 2, 3, 4];
> 
> instead, but if you're not dealing with an initializaton or assignment like
> this (e.g. you're passing the array to a functon), then the cast is the way
> you do it. Normally, you do it with integer literals, and I could see an
> argument that it shouldn't allow it without VRP being used to make it work,
> but it _is_ a cast, and casts are a bit of a blunt instrument.
> 
> So, I really don't think that it's a bug.
> 

It's quite possible that you aren't understanding what is happening:

ubyte[] arr = cast(ubyte[])[555];
writeln(arr); // [43]

Why is this not a bug? I didn't cast the 555 to a ubyte, so it should either complain that it can't do it, or give me an array of 4 bytes.

I guess it could be explained as the same thing as:

ubyte[] arr = [cast(ubyte)555];

But this is surprisingly weird behavior.

-Steve

On Sunday, March 18, 2018 14:56:04 Steven Schveighoffer via Digitalmars-d wrote:
> On 3/18/18 2:24 PM, Jonathan M Davis wrote:
> > On Sunday, March 18, 2018 13:10:28 Steven Schveighoffer via Digitalmars-d
> >
> > wrote:
> >> On 3/18/18 4:34 AM, sdvcn wrote:
> >>> dchar v11=dchar.max;
> >>>
> >>>       auto vp11 = [v11];
> >>>
> >>>       auto v2 = cast(ubyte[]) (vp11);   //v2.length=4
> >>>       auto v22 = cast(ubyte[])( [v11]); //v2.length=1
> >>
> >> This seems like a bug to me.
> >>
> >> It appears that v22 has truncated v11 to a byte and made only a single byte array out of it.
> >
> > Except that that's precisely how you usually get an array any integral type smaller than an integer. e.g.
> >
> > auto arr = cast(ubyte[])([1, 2, 3, 4]);
> >
> > In this case, you could do
> >
> > ubyte[] arr = [1, 2, 3, 4];
> >
> > instead, but if you're not dealing with an initializaton or assignment like this (e.g. you're passing the array to a functon), then the cast is the way you do it. Normally, you do it with integer literals, and I could see an argument that it shouldn't allow it without VRP being used to make it work, but it _is_ a cast, and casts are a bit of a blunt instrument.
> >
> > So, I really don't think that it's a bug.
>
> It's quite possible that you aren't understanding what is happening:
>
> ubyte[] arr = cast(ubyte[])[555];
> writeln(arr); // [43]
>
> Why is this not a bug? I didn't cast the 555 to a ubyte, so it should either complain that it can't do it, or give me an array of 4 bytes.
>
> I guess it could be explained as the same thing as:
>
> ubyte[] arr = [cast(ubyte)555];
>
> But this is surprisingly weird behavior.

That's exactly what it's doing, and when you have multiple elements in the literal, it quickly gets a lot more pleasant than casting each element individually. e.g.

cast(ubyte[])[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10]

vs

[cast(ubyte)0, cast(ubyte)1, cast(ubyte)2, cast(ubyte)3, cast(ubyte)4,
 cast(ubyte)5, cast(ubyte)6, cast(ubyte)7, cast(ubyte)8, cast(ubyte)9,
 cast(ubyte)10]

I use this trick all the time when creating arrays of integral types smaller than int, precisely because casting each element is a royal pain and way harder to read.

- Jonathan M Davis

On Sunday, 18 March 2018 at 20:07:28 UTC, Jonathan M Davis wrote:
> On Sunday, March 18, 2018 14:56:04 Steven Schveighoffer via Digitalmars-d wrote:
>> On 3/18/18 2:24 PM, Jonathan M Davis wrote:
>> > On Sunday, March 18, 2018 13:10:28 Steven Schveighoffer via Digitalmars-d
>> >
>> > wrote:
>> >> On 3/18/18 4:34 AM, sdvcn wrote:
>> >>> dchar v11=dchar.max;
>> >>>
>> >>>       auto vp11 = [v11];
>> >>>
>> >>>       auto v2 = cast(ubyte[]) (vp11);   //v2.length=4
>> >>>       auto v22 = cast(ubyte[])( [v11]); //v2.length=1
>> >>
>> >> This seems like a bug to me.
>> >>
>> >> It appears that v22 has truncated v11 to a byte and made only a single byte array out of it.
>> >
>> > Except that that's precisely how you usually get an array any integral type smaller than an integer. e.g.
>> >
>> > auto arr = cast(ubyte[])([1, 2, 3, 4]);
>> >
>> > In this case, you could do
>> >
>> > ubyte[] arr = [1, 2, 3, 4];
>> >
>> > instead, but if you're not dealing with an initializaton or assignment like this (e.g. you're passing the array to a functon), then the cast is the way you do it. Normally, you do it with integer literals, and I could see an argument that it shouldn't allow it without VRP being used to make it work, but it _is_ a cast, and casts are a bit of a blunt instrument.
>> >
>> > So, I really don't think that it's a bug.
>>
>> It's quite possible that you aren't understanding what is happening:
>>
>> ubyte[] arr = cast(ubyte[])[555];
>> writeln(arr); // [43]
>>
>> Why is this not a bug? I didn't cast the 555 to a ubyte, so it should either complain that it can't do it, or give me an array of 4 bytes.
>>
>> I guess it could be explained as the same thing as:
>>
>> ubyte[] arr = [cast(ubyte)555];
>>
>> But this is surprisingly weird behavior.
>
> That's exactly what it's doing, and when you have multiple elements in the literal, it quickly gets a lot more pleasant than casting each element individually. e.g.
>
> cast(ubyte[])[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10]
>
> vs
>
> [cast(ubyte)0, cast(ubyte)1, cast(ubyte)2, cast(ubyte)3, cast(ubyte)4,
>  cast(ubyte)5, cast(ubyte)6, cast(ubyte)7, cast(ubyte)8, cast(ubyte)9,
>  cast(ubyte)10]
>
> I use this trick all the time when creating arrays of integral types smaller than int, precisely because casting each element is a royal pain and way harder to read.
>
> - Jonathan M Davis



代码目的是转换任意类行到ubyte[]
The object of the code is to convert any class of rows to ubyte[]

但是非数组类无法正确转化
But non - array classes can't be converted correctly



void pByte(T)(T v)
{
	ubyte[] uBuf;
	static if(isArray!T){
		uBuf = cast(ubyte[])mKey;
	}else{
		uBuf = cast(ubyte[])[v];
	}
	writeln(uBuf);
}

无法得到正确结果
Unable to get the correct result

On 3/18/18 4:07 PM, Jonathan M Davis wrote:

> That's exactly what it's doing, and when you have multiple elements in the
> literal, it quickly gets a lot more pleasant than casting each element
> individually. e.g.
> 
> cast(ubyte[])[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10]
> 
> vs
> 
> [cast(ubyte)0, cast(ubyte)1, cast(ubyte)2, cast(ubyte)3, cast(ubyte)4,
>   cast(ubyte)5, cast(ubyte)6, cast(ubyte)7, cast(ubyte)8, cast(ubyte)9,
>   cast(ubyte)10]
> 
> I use this trick all the time when creating arrays of integral types smaller
> than int, precisely because casting each element is a royal pain and way
> harder to read.

Let me adjust your example a bit, and see if you still agree:

auto bytes = cast(ubyte[])[55_444, 289, 1_000_000, 846, 123_456_789];

writeln(bytes); // [148, 33, 64, 78, 21]

I have used cast(ubyte[]) to get ubytes as well, but I normally would do this for values that actually *could be* ubytes. for values higher than ubytes, I would not have expected implicit truncation. It's especially confusing to someone who has seen when you cast an int[] to a ubyte[], and gets the bytes for that same data. When I use cast(ubyte[]), I took it to mean "pretend this is a ubyte[] literal", not "cast each element to ubyte".

I can also see this biting someone who has a long set of ubytes, and accidentally does one that is larger than 255.

-Steve

On Monday, 19 March 2018 at 11:20:05 UTC, Steven Schveighoffer wrote:
> On 3/18/18 4:07 PM, Jonathan M Davis wrote:
>
>> That's exactly what it's doing, and when you have multiple elements in the
>> literal, it quickly gets a lot more pleasant than casting each element
>> individually. e.g.
>> 
>> cast(ubyte[])[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10]
>> 
>> vs
>> 
>> [cast(ubyte)0, cast(ubyte)1, cast(ubyte)2, cast(ubyte)3, cast(ubyte)4,
>>   cast(ubyte)5, cast(ubyte)6, cast(ubyte)7, cast(ubyte)8, cast(ubyte)9,
>>   cast(ubyte)10]
>> 
>> I use this trick all the time when creating arrays of integral types smaller
>> than int, precisely because casting each element is a royal pain and way
>> harder to read.
>
> Let me adjust your example a bit, and see if you still agree:
>
> auto bytes = cast(ubyte[])[55_444, 289, 1_000_000, 846, 123_456_789];
>
> writeln(bytes); // [148, 33, 64, 78, 21]
>
> I have used cast(ubyte[]) to get ubytes as well, but I normally would do this for values that actually *could be* ubytes. for values higher than ubytes, I would not have expected implicit truncation. It's especially confusing to someone who has seen when you cast an int[] to a ubyte[], and gets the bytes for that same data. When I use cast(ubyte[]), I took it to mean "pretend this is a ubyte[] literal", not "cast each element to ubyte".
>
> I can also see this biting someone who has a long set of ubytes, and accidentally does one that is larger than 255.
>
> -Steve

An even funnier example.

    auto bytes = cast(ubyte[])cast(int[])[55_444, 289, 1_000_000, 846, 123_456_789];
    auto bytes2 = cast(int[])[55_444, 289, 1_000_000, 846, 123_456_789];
    auto bytes3 = cast(ubyte[])bytes2;

    writeln(bytes); // Guess what the output is here.
    writeln(bytes2); // Prints:[55444, 289, 1000000, 846, 123456789]
    writeln(bytes3); // Prints: [148, 216, 0, 0, 33, 1, 0, 0, 64, 66, 15, 0, 78, 3, 0, 0, 21, 205, 91, 7]

On Monday, 19 March 2018 at 11:20:05 UTC, Steven Schveighoffer wrote:
> Let me adjust your example a bit, and see if you still agree:
>
> auto bytes = cast(ubyte[])[55_444, 289, 1_000_000, 846, 123_456_789];
>
> writeln(bytes); // [148, 33, 64, 78, 21]
>
> I have used cast(ubyte[]) to get ubytes as well, but I normally would do this for values that actually *could be* ubytes. for values higher than ubytes, I would not have expected implicit truncation. It's especially confusing to someone who has seen when you cast an int[] to a ubyte[], and gets the bytes for that same data. When I use cast(ubyte[]), I took it to mean "pretend this is a ubyte[] literal", not "cast each element to ubyte".
>
> I can also see this biting someone who has a long set of ubytes, and accidentally does one that is larger than 255.
>
> -Steve

Raw `cast` is just nasty. It's overloaded and confusing. Wrapper template functions like `reinterpretBitsAs` can help alleviate the pain, e.g. `assert([1, 2, 3].reinterpretBitsAs!(ubyte[]).length == 12);`.

I feel like C++ got it right (or just less wrong) with casts.