Issue 8409 - Proposal: implement arr.dup in library

[9rnsr]

http://d.puremagic.com/issues/show_bug.cgi?id=8409

This works with UFCS, and doesn't break existing codes.

arr.dup, arr.idup	built-in property is used
arr.dup()	library version is used

Requires: dlang/dmd#1110

[alexrp]

Maybe nothrow too?

[9rnsr]

Ah..., yes. new can become nothrow. Thanks.

[9rnsr]

Apply two fixes from @alexp's comments.

[tgehr]

You might want to fix this:

struct S{
immutable int x;
}
void main(){
S[] x = [];
auto y=x.dup();
}

[9rnsr]

Added test case.

[andralex]

Is X technically needed, or just for organizational purposes? In the latter case I think the function is small enough to just delete it.

[tgehr]

It is 'needed' because the grammar on TDPL (which would, as far as I can see, not allow local struct definitions either), as well as DMD's parser are all messed up. It is not currently possible to declare a local template without these tricks, because parsing fails.

[andralex]

This should be filed as a bug. It breaks the TOWD principle.

[MartinNowak]

What's the TOWD principle?

[alexrp]

This seems to be failing the auto tester?

[9rnsr]

Because this is based on the compiler fix (dlang/dmd#1110), so until merging it, auto tester will not become green.

[alexrp]

Are we just waiting on review of the compiler pull request in order to move forward with this? (I.e. does everyone agree that this is a good idea?)

[alexrp]

@9rnsr see my previous comment

[MartinNowak]

How about replacing the built-in properties instead of adding an alternative syntax?

[alexrp]

The DMD pull request is merged now. Can we merge this?

[ghost]

This is really odd, we already have the rule in D that foo and foo() are equivalent for functions, but now we're assigning a special meaning to it just for .dup. I think this is a step in the wrong direction..

[9rnsr]

Thanks many comments. I need more thought about this. Temporarily close.