[phobos] slice vs. append

[Andrei Alexandrescu]

On 03/22/2010 10:38 AM, Steve Schveighoffer wrote:
> ----- Original Message ----
>> From: Andrei Alexandrescu<andrei at erdani.com> Yes, but at the end of
>> the day, you do have a primitive that tells you whether a given
>> array extends to the end of a given block and consequently is ok to
>> append to in-place. Correct?
>
> The cache is not that.  That primitive is determined by the GC block
> info.  Essentially, with the blockinfo, I can tell where an array
> ends, and therefore know whether it can be appended in place.  This
> is true regardless of whether the blockinfo is in the cache or not.
>
> The cache simply allows me to skip the expensive blockinfo lookup.
> Essentially, by removing a blockinfo from the cache, it doesn't
> prevent any behavior, it just makes it slower.

OK.

>> You told me that for best results I need to do:
>>
>> arr.length = 0; arr.assumeSafeAppend();
>>
>> The last call makes the system assume that I saved no subarray of
>> arr that I'm interested in keeping. No?
>
> Yes.  I think I see where you were going with the request.  You want
> to remove that second line?  It simply does not work, because an
> array isn't marked differently because it was sliced, and the runtime
> cannot tell.
>
> What about code like this:
>
> auto arr2 = arr; arr2.length = 0;

Yeah, it looks like array assignment also needs to be hooked.

> Now, if you assume because the user used length instead of slicing,
> it's ok to append, what happens to arr?  That is why we need the
> assume call.  To get what you want, any time an array is copied, even
> via function calls, you'd have to kill appending to it.  I don't
> think that it's worth the performance loss on the common operations
> in order to save a line of code in an uncommon situation.

I agree.


Andrei