• AlbertG (53/53) Jul 14 This week I have completed the transition from `_adEq2` to

AlbertG <albert.guiman protonmail.com> writes:

This week I have completed the transition from `_adEq2` to 
`__equals` + `memcmp`[1] for array comparison.

There are a few changes in how the lowering is done:

- Comparison between arrays of the same struct type, when the 
structs can be directly compared using `memcmp`:

Example with dynamic arrays:
```d
struct A {
     char[256] x = 1;

     this(this){}
}
A[] a = new A[2];
A[] b = new A[2];
assert(a == b);
```
Before: lowered to `__equals`(which did not use `memcmp`).
Now: lowered to `memcmp` directly.

Example with static arrays:
```d
struct A {
     char[256] x = 1;

     this(this){}
}
A[2] a = new A[2];
A[2] b = new A[2];
assert(a == b);
```
Before: lowered to `_adEq2`.
Now: lowered to `memcmp` directly.

- Comparison between static arrays, with elements that cannot be 
compared using `memcmp`:

Examples:
```d
double[2] a = new double[2];
double[2] b = new double[2];
assert(a == b);
```

```d
class A {
     int a;

     this(){ this.a = 1; }
}

A[2] a = new A[2];
A[2] b = new A[2];
assert(a == b);
```

Before: lowered to `_adEq2`.
Now: lowered to `__equals`.

I have also noticed a slight reduction in `libphobos` and 
`libdruntime` sizes (around 0.5% when compiled with dmd), which 
is probably due to the attribute stripping that occurs before the 
lowering to `__equals` is performed.

[1] https://github.com/dlang/dmd/pull/21513