Rudi Grinberg

We're actually trying to get rid of contexts from the engine and make them strictly rule only, so this PR would make it more more difficult for us. I think...

> @rgrinberg is [63f17d1](https://github.com/ocaml/dune/commit/63f17d133763693cfd77c048ae7aa960c136554a) what you had in mind? I tried moving the `with-directory` annotation function to `User_message` to simplify things further, but this creates a cycle between `User_message` and...

That's a bit of a pain. I think you can just use a string path though.

@pmwhite can you check if this change is going to work internally?

Ok, thanks. In general, I'm quite happy to get rid of this "annots" map and just shove things directly into the user error type. Once the internal patch is merged,...

This should be same as https://github.com/ocaml/dune/issues/9700 as well - but on the lock file generation side. We're currently generating `file://`, but it's buggy. So we need to make it robust...

Good point. Why are we even recording the build command for pins again? If we have the checksum for the source, that already fixes the build command. If we don't...

@bmillwood What you're suggesting sounds a lot like ocamlbuild's tags system. I don't have many fond memories of ocamlbuild, but I thought that how it let one concisely specify where...

While this feature is important, I'd actually recommend for the code generators to change to output correct warning removing annotations (`[@@@ocaml.warning ".."]`). This will work for all build systems, and...

By the way, the way that I'm currently thinking of solving this problem is allow glob/file constructors in the flags field. This will be consistent with how we'd like to...