RFC: naming for FrontTransversal and Transversal ranges

[Bill Baxter]

On Fri, May 1, 2009 at 6:25 PM, Andrei Alexandrescu
<SeeWebsiteForEmail at erdani.org> wrote:
> Bill Baxter wrote:
>>
>> On Fri, May 1, 2009 at 4:23 PM, Andrei Alexandrescu
>> <SeeWebsiteForEmail at erdani.org> wrote:
>>>
>>> I guess at that point the would-be D user would be entitled to make me a
>>> lavaliere out of my Matrix library and move on.
>>
>> Python, and therefore NumPy, are reference based.
>
> In a language that's reference based, things can't be helped. They can be
> helped in D though. I wouldn't want to later cringe that I missed the
> opportunity.
>
>> So far I haven't seen any scientists posting on the list about how
>> having their arrays and matrices be references is driving them crazy.
>> It may be surprising to some of them at first, but even non-hard-core
>> coders seem to be able to handle it.
>
> To me this sounds more like an opportunity to do the right thing.

Could be.  I'm not arguing that refs are best, just giving some
evidence from the NumPy community to counter your assertion that
scientists would be confused by reference-based array and matrix
types.


>> It helps that dummy assignments
>> like a = b are rare.   More often you have things like a = b + c, and
>> that creates a new matrix.  Or  a += b, which pretty clearly mutates
>> a.
>
> Oh rly. How about this then?
>
> Matrix a, b, c;
> ...
> c = a;
> a += b;
>
> Does the last operation mutate c as well?

I said "assignments like a = b are rare" and you put one of those in
your example.  Yes, when you have an a=b anywhere you've got to pay
attention and make sure you didn't mean a=b.dup.

And the other thing is that when you pass a matrix to a function, you
need to be aware of whether the function mutates that matrix or not.
That's really the biggest pain, and source of hardest to find bugs.

Matlab goes the other extreme and makes everything value semantics.
It does a lot of COW optimizations under the hood to make it tolerably
efficient even when tossing around big matrices.  But it's annoying if
you want to write a function to make a small update to an existing
matrix.  You have no choice but to return a copy and hope that
Matlab's optimizer is smart enough to eliminate the unnecessary copy.

I agree with the poster below that what I usually want is value
semantics with small values and matrices, reference semantics with big
matrices.

--bb