[Mlir-commits] [mlir] [MLIR][CAPI][Python] Add support for querying memory effect instances (PR #213459)

[llvmlistbot at llvm.org]

PragmaTwice wrote:

> Does this sound right to you?

Yah I think you are right. In the C API, for consuming a list we can just pass `size_t len, MlirSomeValue *values` as arguments. But for producing a list, it become more complicated due to lifetime issues, and the existing pattern is to pass a callback with `size_t len, MlirSomeValue *values` as parameters. The callback can be treated as a "scope" so the lifetime can be a bit easier. I've added a descrption about this callback type to PR description : )

@adam-smnk @rolfmorel Would you like to comment on changes in the Python side? Previously `effects` is a mutable input argument, and these functions will append new effects into effects list (`MemoryEffectInstancesList`), e.g.

```python
    @staticmethod
    def get_effects(op: ir.Operation, effects):
        transform.only_reads_handle(op.op_operands, effects)
        transform.produces_handle(op.results, effects)
        transform.only_reads_payload(effects)
```
now it will return a new Python-native `list` and we can use `+` to concat lists and return.
```python
    def get_effects(op: ir.Operation):
        return (
            transform.only_reads_handle(op.op_operands)
            + transform.produces_handle(op.results)
            + transform.only_reads_payload()
        )
```

https://github.com/llvm/llvm-project/pull/213459