http://dlang.org/phobos/std_bitmanip.html

Does anyone know how to bit-shift a BitArray?

I'm trying to make spikes in a neural network travel along the
bits as they have various lengths.

On 12/12/14 8:39 AM, Trollgeir wrote:
> http://dlang.org/phobos/std_bitmanip.html
>
> Does anyone know how to bit-shift a BitArray?
>
> I'm trying to make spikes in a neural network travel along the
> bits as they have various lengths.

That is a surprising omission from a bit-oriented type...

You also cannot opSlice a BitArray. Nice for enhancement requests, if you want to add them to the issue tracker.

-Steve

On Fri, Dec 12, 2014 at 11:13:38AM -0500, Steven Schveighoffer via Digitalmars-d-learn wrote:
> On 12/12/14 8:39 AM, Trollgeir wrote:
> >http://dlang.org/phobos/std_bitmanip.html
> >
> >Does anyone know how to bit-shift a BitArray?
> >
> >I'm trying to make spikes in a neural network travel along the bits as they have various lengths.
> 
> That is a surprising omission from a bit-oriented type...
> 
> You also cannot opSlice a BitArray. Nice for enhancement requests, if you want to add them to the issue tracker.
[...]

I've started working on an implementation of this... but it's not very clear what the correct semantics should be. For example, if my starting BitArray b is 1101, say, what should be the result after b>>=1? Should it be 0110, 110, or 01101?


T

-- 
It is widely believed that reinventing the wheel is a waste of time; but I disagree: without wheel reinventers, we would be still be stuck with wooden horse-cart wheels.

On 12/12/14 2:17 PM, H. S. Teoh via Digitalmars-d-learn wrote:
> On Fri, Dec 12, 2014 at 11:13:38AM -0500, Steven Schveighoffer via Digitalmars-d-learn wrote:
>> On 12/12/14 8:39 AM, Trollgeir wrote:
>>> http://dlang.org/phobos/std_bitmanip.html
>>>
>>> Does anyone know how to bit-shift a BitArray?
>>>
>>> I'm trying to make spikes in a neural network travel along the
>>> bits as they have various lengths.
>>
>> That is a surprising omission from a bit-oriented type...
>>
>> You also cannot opSlice a BitArray. Nice for enhancement requests, if
>> you want to add them to the issue tracker.
> [...]
>
> I've started working on an implementation of this... but it's not very
> clear what the correct semantics should be. For example, if my starting
> BitArray b is 1101, say, what should be the result after b>>=1? Should
> it be 0110, 110, or 01101?

0110

In other words, I would assume the same semantics as an unsigned int.

In other other words, it's like each bit moves one to the right, and the bit that has no source gets a 0.

It may be useful to add some other functions, such as roll, which would move bits that fall off the end back onto the top.

And maybe expose shift as a function, which allows you to specify the bit value (or maybe range of bits) that should be shifted in from the other side.

Note, I would not allow just opBinary, only opBinaryAssign. No reason to construct temporary BitArrays.

-Steve

Here's my implementation of <<= and >>= for BitArray:

	https://github.com/D-Programming-Language/phobos/pull/2797

While working with the code, I found that there are a lot of areas that need improvement. If I have some time I'll file separate PR's for them.


T

-- 
VI = Visual Irritation

On Friday, 12 December 2014 at 19:35:26 UTC, Steven Schveighoffer wrote:
> On 12/12/14 2:17 PM, H. S. Teoh via Digitalmars-d-learn wrote:
>> On Fri, Dec 12, 2014 at 11:13:38AM -0500, Steven Schveighoffer via Digitalmars-d-learn wrote:
>>> On 12/12/14 8:39 AM, Trollgeir wrote:
>>>> http://dlang.org/phobos/std_bitmanip.html
>>>>
>>>> Does anyone know how to bit-shift a BitArray?
>>>>
>>>> I'm trying to make spikes in a neural network travel along the
>>>> bits as they have various lengths.
>>>
>>> That is a surprising omission from a bit-oriented type...
>>>
>>> You also cannot opSlice a BitArray. Nice for enhancement requests, if
>>> you want to add them to the issue tracker.
>> [...]
>>
>> I've started working on an implementation of this... but it's not very
>> clear what the correct semantics should be. For example, if my starting
>> BitArray b is 1101, say, what should be the result after b>>=1? Should
>> it be 0110, 110, or 01101?
>
> 0110
>
> In other words, I would assume the same semantics as an unsigned int.
>
> In other other words, it's like each bit moves one to the right, and the bit that has no source gets a 0.
>

There's a dedicated >>> operator for unsigned right shift, the normal >> operator does a signed right shift, i.e. it copies the left-most bit:
http://dlang.org/expression#ShiftExpression

IMO, for consistency, bitarray should behave the same.

> It may be useful to add some other functions, such as roll, which would move bits that fall off the end back onto the top.

Yes, that would be useful.

On 12/13/14 5:47 AM, "Marc =?UTF-8?B?U2Now7x0eiI=?= <schuetzm@gmx.net>" wrote:
> On Friday, 12 December 2014 at 19:35:26 UTC, Steven Schveighoffer wrote:
>> On 12/12/14 2:17 PM, H. S. Teoh via Digitalmars-d-learn wrote:
>>> I've started working on an implementation of this... but it's not very
>>> clear what the correct semantics should be. For example, if my starting
>>> BitArray b is 1101, say, what should be the result after b>>=1? Should
>>> it be 0110, 110, or 01101?
>>
>> 0110
>>
>> In other words, I would assume the same semantics as an unsigned int.
>>
>> In other other words, it's like each bit moves one to the right, and
>> the bit that has no source gets a 0.
>
> There's a dedicated >>> operator for unsigned right shift, the normal >>
> operator does a signed right shift, i.e. it copies the left-most bit:
> http://dlang.org/expression#ShiftExpression
>
> IMO, for consistency, bitarray should behave the same.

But BitArray is not signed, it's an array of bits, none of which are signed or designated as the sign bit. The unsigned shift operator is only for signed integral types, for unsigned values >> is simply a shift of bits. Note that your assertion that >> "copies the left-most bit" is not in the text, and is incorrect. It copies the sign bit, which doesn't exist in an unsigned type.

I think >> and >>> should do the same thing, and be unsigned shifts on BitArray.

-Steve

On Monday, 15 December 2014 at 15:19:25 UTC, Steven Schveighoffer wrote:
> On 12/13/14 5:47 AM, "Marc =?UTF-8?B?U2Now7x0eiI=?= <schuetzm@gmx.net>" wrote:
>> On Friday, 12 December 2014 at 19:35:26 UTC, Steven Schveighoffer wrote:
>>> On 12/12/14 2:17 PM, H. S. Teoh via Digitalmars-d-learn wrote:
>>>> I've started working on an implementation of this... but it's not very
>>>> clear what the correct semantics should be. For example, if my starting
>>>> BitArray b is 1101, say, what should be the result after b>>=1? Should
>>>> it be 0110, 110, or 01101?
>>>
>>> 0110
>>>
>>> In other words, I would assume the same semantics as an unsigned int.
>>>
>>> In other other words, it's like each bit moves one to the right, and
>>> the bit that has no source gets a 0.
>>
>> There's a dedicated >>> operator for unsigned right shift, the normal >>
>> operator does a signed right shift, i.e. it copies the left-most bit:
>> http://dlang.org/expression#ShiftExpression
>>
>> IMO, for consistency, bitarray should behave the same.
>
> But BitArray is not signed, it's an array of bits, none of which are signed or designated as the sign bit. The unsigned shift operator is only for signed integral types, for unsigned values >> is simply a shift of bits. Note that your assertion that >> "copies the left-most bit" is not in the text, and is incorrect. It copies the sign bit, which doesn't exist in an unsigned type.
>
> I think >> and >>> should do the same thing, and be unsigned shifts on BitArray.

Yes, yebblies pointed that out in the PR. I was confused, because the documentation doesn't say that `>>` and `>>>` only behave differently for signed integers.