Leonardo Yvens

I imagine that if we had a dependency graph for goals, we could have rustc error messages only report the minimal set of obligation failures that need to be fixed...

I approve this and any future relicensing which may require approval.

Could the issue be better described, is this a correctness fix or an optimization? If it's correctness I'm surprised that `match_and_decode` would somehow be giving false positives.

Thanks for the explanation, please wait for a review from me before merging.

On the correctness fix, the question is why is a PoI being written in blocks that processed no triggers. I believe this is due to this `start_handler` call: https://github.com/graphprotocol/graph-node/blob/4850b828f9257a7725758e5dc3d168b39305a790/core/src/subgraph/trigger_processor.rs#L37-L39 Which...

> The main issue is the assumption that any block then went through filtering will have at least a single valid trigger. True this bug exposes this accidental assumption, but...

I'm neutral on whether grouped imports are nicer or not. But running this manually every now and then would cause a bunch of merge conflicts with any open PRs, so...

We have `with_conn` now so the API question is solved. Maybe we don't use it everywhere we should but keeping this issue open isn't what's going to solve that.

Actually we have way too much `spawn_blocking` around our code to consider this done. This will be done when `graph::task_spawn::spawn_blocking` is no longer in use, since it assumes that a...

Sounds fun, we could allow a free-form yaml map which is converted into a context.