Add ability to compare const/immutable classes

[dlangBugzillaToGithub]

Denis Shelomovskii (@denis-sh) reported this on 2013-11-19T13:08:50Z

Transferred from https://issues.dlang.org/show_bug.cgi?id=11557

Description

Currently because of Issue 1824 `opEquals` and `opCmp` aren't `const`. It's bad but works if these functions are implemented carefully.

Current dmd rejects `opCmp` call:
---
const Object o1, o2;
assert(o1 == o2); // ok, calls `opEquals`
assert(o1 <= o2); // Error: mutable method object.Object.opCmp is not callable using a const object
---

As a result e.g. `std.typecons.Tuple`-s with classes don't have `opCmp` and fails with "TypeInfo.compare is not implemented" `Error` if used as an associative array key. Error at runtime is really nasty.

[dlangBugzillaToGithub]

verylonglogin.reg commented on 2013-11-19T13:14:09Z

A workaround for `std.typecons.Tuple` regression:
https://github.com/D-Programming-Language/phobos/pull/1707

[dlangBugzillaToGithub]

k.hara.pg commented on 2013-11-20T01:26:06Z

(In reply to comment #0)
> Currently because of Issue 1824 `opEquals` and `opCmp` aren't `const`. It's bad
> but works if these functions are implemented carefully.
> 
> Current dmd rejects `opCmp` call:
> ---
> const Object o1, o2;
> assert(o1 == o2); // ok, calls `opEquals`
> assert(o1 <= o2); // Error: mutable method object.Object.opCmp is not callable
> using a const object
> ---
> 
> As a result e.g. `std.typecons.Tuple`-s with classes don't have `opCmp` and
> fails with "TypeInfo.compare is not implemented" `Error` if used as an
> associative array key. Error at runtime is really nasty.

Why this issue is marked as "regression"? As far as I know, the code you shown had never worked correctly.

[dlangBugzillaToGithub]

verylonglogin.reg commented on 2013-11-20T07:31:13Z

(In reply to comment #2)
> Why this issue is marked as "regression"? As far as I know, the code you shown
> had never worked correctly.

Associative arrays use `TypeInfo.compare` and `std.typecons.Tuple` has its `opCmp` with by element comparison for a long time. As `Tuple`-s stopped working as AA keys (it worked at least a year ago) I decided it was a compiler change. If it's not please remove the REGRESSION status.

[dlangBugzillaToGithub]

k.hara.pg commented on 2013-11-21T07:30:09Z

(In reply to comment #3)
> (In reply to comment #2)
> > Why this issue is marked as "regression"? As far as I know, the code you shown
> > had never worked correctly.
> 
> Associative arrays use `TypeInfo.compare` and `std.typecons.Tuple` has its
> `opCmp` with by element comparison for a long time. As `Tuple`-s stopped
> working as AA keys (it worked at least a year ago) I decided it was a compiler
> change. If it's not please remove the REGRESSION status.

See my comment in your PR. To me the old worked behavior was essentially wrong, and I think it should be kept broken until class const correctness will be fixed.

https://github.com/D-Programming-Language/phobos/pull/1707#issuecomment-28992632

[dlangBugzillaToGithub]

verylonglogin.reg commented on 2013-11-24T00:56:24Z

Filed Issue 11588 for inconsistency in abilities to compare `const`/`immutable` classes and Issue 11591 for Phobos regression.

This issue is a request to add ability to compare `const`/`immutable` classes just like currently allowed testing for equality. It will fix Issue 11588 if accepted.