delete! semantics for vector vs. set

[mlubin]

There's an issue with the overloaded use of delete! to mean removing both a particular index from a vector and a particular element from a set: vectors partially implement set functions, and some set functions return vectors. See below.

julia> s1 = IntSet(10,12,13);s2 = IntSet(14,15,16);s3 = IntSet();

julia> u = union(union(s1,s2),s3)
IntSet(10, 12, 13, 14, 15, 16)

julia> delete!(u,12)
12

julia> u2 = union(s1,s2,s3)
6-element Any Array:
 10
 12
 13
 14
 15
 16

julia> delete!(u2,12)
ERROR: BoundsError()
 in delete! at array.jl:765

This particular case would be fixed by implementing union(s::IntSet,sets::IntSet...) to return an IntSet, but I think the problem is a bit bigger than this.

[JeffBezanson]

How about delindex!?

[mlubin]

Sounds good to me. delindex! for deleting indices in a vector and keys in a dictionary, and delete! for deleting elements in a set.

delete! could be defined for vectors and dictionaries as removing matching elements as well, but maybe delete!(::Dict,key) and delete!(::Array,i) should be deprecated for a while before suddenly changing the meaning of the function.

[StefanKarpinski]

Since we probably want the same function for vectors and hashes, this needs a little more thought. Maybe deletekey!?

[mlubin]

delindex! is a bit awkward for dicts and deletekey! is a bit awkward for vectors. Maybe they should be separate?

[mlubin]

The semantics are also different. After removing a key k from dict d, you can no longer access d[k]. For vectors, the elements are shifted, and the index you deleted could still be valid.

[StefanKarpinski]

Good point about the rather different semantics. This is all part of the final collections API bikeshedding push we need to do.

[JeffBezanson]

Stefan and I decided we will rename delete! for Vector to splice!, and add a form splice!(A, out, in) that replaces the indexes out with the values in, e.g.

julia> A = [1,2,3,4,5,6];

julia> splice!(A, 2:3, [7,8,9])
[2, 3]

julia> A
[1,7,8,9,4,5,6]

[JeffBezanson]

In other words, the default value of the third argument to splice! is an empty array.

[StefanKarpinski]

We could also support variants where the indices to remove/replace are non-contiguous, although that would obviously be more complicated to implement, but that's kind of why it's useful to have, actually.

[StefanKarpinski]

Actually, scratch that since when the range isn't contiguous it's ambiguous where to put the replacement values.

[kmsquire]

While I can see the problem highlighted here, I'm wondering if there is a better fix than the one committed?

One of the things I like about julia is that it combines performance with enough common sense that most things are intuitive and just work. I think this change goes against that. delete!() from an array is a common enough operation, and deleting by index is very intuitive. Finding splice!() when I'm looking to delete an element from an array seems potentially quite frustrating.

[johnmyleswhite]

I agree with Kevin that this is a non-intuitive change of names.

-- John

On May 31, 2013, at 8:36 PM, Kevin Squire notifications@github.com wrote:

While I can see the problem highlighted here, I'm wondering if there is a better fix than the one committed?

One of the things I like about julia is that it combines performance with enough common sense that most things are intuitive and just work. I think this change goes against that. delete!() from an array is a common enough operation, and deleting by index is very intuitive. Finding splice!() when I'm looking to delete an element from an array seems potentially quite frustrating.

—
Reply to this email directly or view it on GitHub.

[mlubin]

I'd agree with @kmsquire that there should be a more intuitive and searchable name for ''splice!'' when two arguments are used.