opUnaryAssign

[Quirin Schroll]

On Wednesday, 3 July 2024 at 13:26:35 UTC, monkyyy wrote:
> On Wednesday, 3 July 2024 at 10:43:14 UTC, Quirin Schroll wrote:
>> Index operators (e.g. `obj[i]`) have special overloads so that 
>> they hook assignment: `obj[i] = rhs` lowers to 
>> `obj.opIndexAssign(rhs, i)`. Why not back-port this to 
>> dereferencing? `*obj` lowers to `obj.opUnary!"*"`, which, to 
>> be assignable, must return by reference. Why not add 
>> `opUnaryAssign(string op)` which can hook, in principle, `+obj 
>> = rhs`, `-obj = rhs`, `~obj = rhs`, `*obj = rhs`, `++obj = 
>> rhs`, and `--obj = rhs` (of which I expect only `*obj = rhs` 
>> to be used regularly).
>
> wouldnt `+obj=rhs` combined with "assignment returns" be 
> ambiguous in some case?

In the [expression 
grammar](https://dlang.org/spec/expression.html), clearly `+obj` 
has precedence over `=`. The precedence table is 
[here](https://wiki.dlang.org/Operator_precedence). The gist is:
1. Postfix operators (includes call `()`, index `[]`, and member 
access `.`)
2. Prefix operators
3. Binary operators

There are exceptions, e.g. binary `!` for templates is top, `^^` 
is stronger than unary prefix, and `=>` has different precedence 
on the left and right, but for the most part, the 3-item list is 
accurate.