Fix issue 11006 - Make sure the same type is used in pointer ops

[LemonBoy]

If both operands are pointers, and the operator is -, the pointers are subtracted and the result is divided by the size of the type pointed to by the operands. It is an error if the pointers point to different types. The type of the result is ptrdiff_t.

[dlang-bot]

Fix	Bugzilla	Description
✓	11006	Subtraction of pointers for void and non-void types compiles
✓	16967	No switch case fallthrough warnings in in/out contracts

[denis-sh]

It's better to also check error message in test.
Also at least druntime source code must be adjusted first (get rid of "Operands must point to the same type" errors) to be able to merge this.

[LemonBoy]

Good idea, I've amended the test case.
I've already sent two PRs for druntime and phobos, we just have to wait for them to get merged.

[WalterWaldron]

Deprecate first

[MartinNowak]

I'd expect this to break a lot of code, maybe rather require types of same size than requiring identity?

[LemonBoy]

Phobos and druntime have been mostly chill as all the errors were due to the use of void* and byte*. Relying on the size of the types still allows the user to mix and match the types which is not what you want in 99% of the cases (and if you really need to do so you're just explicit cast away) as some types may vary between platforms and one may end up enlarging/shrinking one of the types involved and end up with a not so nice compilation error.
I root for following the specification here.

[MartinNowak]

Smells like this could do more than just dealing with void*.

[LemonBoy]

It just brings the types down to a common one, I couldn't find a case where something more than that was being done.

[LemonBoy]

Should this go trough the deprecation phase too? I think this is a bugfix since we stray away from the specification here.

[Geod24]

Should this go trough the deprecation phase too? I think this is a bugfix since we stray away from the specification here.

Doesn't matter if it's a bug fix or not, if it breaks code and there is a sane depreciation path we have to take it. Remember that a single occurrence in an unused code path in a library will cause downstream projects to fail to compile. That's not something acceptable.

[yebblies]

Are pointers with different modifiers handled correctly? eg const(void)* and void*

[LemonBoy]

Sure, the equivalent predicate covers those cases

[JinShil]

Revived at #7332

[LemonBoy]

You're just grabbing the commits and throwing away all the informations about the author, that's not nice.

[JinShil]

You're just grabbing the commits and throwing away all the informations about the author, that's not nice.

Not at all. I've posted links in the old PRs to the new PRs, and from the new PRs to the old PRs. So there is an audit trail that anyone can follow. In terms of the source code itself, there was no copyright or other information about the author to keep, otherwise I would have kept it. If you're concerned about the merge being in your name, revive them yourself.