Steven S. Lyubomirsky

Can we clarify, by the way, what the rule for shape-checking is so we can be consistent? If you have an assignment like `x: Tensor((m, n), dtype) = some_expr`, does...

Closing because the discussion seems to have been settled and the draft spec reflects strong typing.

I very much like the idea of including "ANF" in the name of the mutator. The issue with visiting binding nodes is an interesting one. I don't think it necessarily...

Another weird inconsistency (possibly a bug?): Branch conditions create an error if given an NDArray and work only with an int constant. Note that bool scalar constants created within a...

Are there any docs on TVM's object system outside of the code base? This would be a good thing to clearly specify for users of Relax, especially since we are...

> Regarding None, I'm not sure if there would be a real case that users might want to pass None - this seems an error to me. Wonder how others...

A brief thought I brought up in our meeting but wanted to leave in writing: For the management of passes and when they are applicable, we may want some automatic...

Thank you for the changes. The new unit test and the new comment are both helpful.

@Lunderberg thank you for the review. I've pushed a change to enable a well-formed check for individual Relax functions (this exposed some bugs in our tests!) and PrimFuncs, which can...

I've determined that the segfault in the tests comes from `tests/python/tir-transform/test_transform_inject_rolling_buffer.py`. According to the backtrace, this comes from checking the StructInfo for a global variable, but oddly, it does not...