lower array equals only if it needs code gen

[WalterBright]

No description provided.

[dlang-bot]

Thanks for your pull request, @WalterBright!

Bugzilla references

Auto-close	Bugzilla	Severity	Description
✗	23771	enhancement	Array equals should be handled by CTFE

Testing this PR locally

If you don't have a local development environment setup, you can use Digger to test this PR:

dub run digger -- build "master + dmd#14981"

[WalterBright]

Fixing this is blocked by https://issues.dlang.org/show_bug.cgi?id=23771

[RazvanN7]

I have rebased to latest master and fixed the failing tests. It turns out we only needed to fix the failing error message.

[RazvanN7]

This cannot be pulled unless #11212 is reverted (cc @kinke ). Prior to #11212 , before lowering, the compiler would do some type checks to make sure that the comparison may be performed (and output errors in case it was not possible). The checks were duplicated in the hook, so the PR made sense because the hook would get lowered regardless if the scope needs codegen or not. However, this PR makes it so that the lowering is performed only in codegen'ed scopes so for situations such as __traits(compiles, arr1 == arr2) you are bypassing and, implicitly, the associated type checks.

[RazvanN7]

The 3 options we have seem to be:

• Drop this PR, always perform the lowering
• Revert Get rid of obsolete __ArrayEq lowering #11212 and get this PR in
• Keep Get rid of obsolete __ArrayEq lowering #11212 and check if the lowering is performed in speculative contexts just mark it as not needing codegen (although this might not interact well with the caching of template instances).

[kinke]

I don't think the revert is a real option, unless you want to re-add __ArrayEq to druntime too, and fix the other issue that my PR apparently fixed.

Adding a EqualExp.lowering field could be one option, and ignoring it for CTFE.

[RazvanN7]

Closing, as this is not the right approach.