There are pathological cases (see here) that might require us to re-compute the codegen plan after the execution of each code generation unit.
We could start off with a pretty naive implementation (i.e. cargo generate-lockfile + re-compute the codegen plan - the units we already generated) and then make it more sophisticated over time (e.g. detect if the newly generated manifest has added any new dependency edges and skip re-generation if that's not the case).
There are pathological cases (see here) that might require us to re-compute the codegen plan after the execution of each code generation unit.
We could start off with a pretty naive implementation (i.e.
cargo generate-lockfile+ re-compute the codegen plan - the units we already generated) and then make it more sophisticated over time (e.g. detect if the newly generated manifest has added any new dependency edges and skip re-generation if that's not the case).