[llvm] [TableGen] Rework error reporting for duplicate Feature/Processor. (PR #102257)

[Sergei Barannikov via llvm-commits]

s-barannikov wrote:

> Also note that this is not true. The records vended out by getAllDerivedDefinitions() are sorted by name and not their lexical order, so even in the previous case the "Previous" definition could have been a later one.

I agree with this.

```
def ProcessorA : ProcessorModel<"SameName", NoSchedModel, []>;

multiclass M {
  def C : ProcessorModel<"SameName", NoSchedModel, []>;
}

def ProcessorB : ProcessorModel<"SameName", NoSchedModel, []>;

defm : M;
```

The old code complained about ProcessorA.
This is also an example where the "source location" is not clearly defined. I would argue that the error should complain about ProcessorB, not about C instantiated from M even though C precedes ProcessorB in source order.


https://github.com/llvm/llvm-project/pull/102257