Stepan Koltsov

Yes. Errors look like this: ``` Code: 33. DB::Exception: Cannot read all data. Bytes read: 3. Bytes expected: 8.: (at row 1)\n: While executing BinaryRowInputFormat. (CANNOT_READ_ALL_DATA) ``` Unhelpful, hard to...

@aherrmann right. Internally we agreed to go with this approach — add `--profile-mode=` to many commands and output profile of whatever is evaluated. So far I have implemented this flag...

I'm not a project member, this comment is FWIW. It should be in a separate crate (like `pingora-prometheus`, `pingora-extras` or something), not optional: nothing in pingora-core depends on prometheus, except...

This change is much larger than it seems. For example, because of how starlark works: `load()` is a statement, not a function. And name resolution is static in starlark: `load(...,...

@izissise buck2 does not build targets or execute commands during target graph construction. There are reasons for that, for example, target determinator, which should be fast and reliable. When build...

@toothbrush7777777 dynamic output allows accessing command output during build time (delayed/dynamic) analysis, but does not allow creation of new targets (or even subtargets). It can be used to sort dependencies...

@izissise I'd suggest instead: - first, describe your use case, and why it cannot be solved with other tools (like for example, by generating `BUCK` files before execution of `buck2`...

> (Weak) rationale for thinking about this: - Opportunity to do optimization: ``` def foo(**kwargs): pass def bar(**kwargs): foo(**kwargs) ``` If kwargs is immutable, we can avoid copying. - Similarly:...

Why not just make fields public? This is mostly just data and I struggle to find examples where it enforces some invariants.

They @nyurik this is breaking change. So it is either: - breaking change - or a codegen option is needed, which would not be useful to many people, because people...