@giordano and @staticfloat this reminded me that we don't record the compat of JLL packages against dependencies all that much. Here I wanted to make sure that I honor the major version compat of MPITrampoline.
Should we not autogenerate a compat for major version for most JLL?

Should we not autogenerate a compat for major version for most JLL?

It's complicated. We generally follow upstream version numbers, but they do not adhere to semver: they can introduce breaking changes in patch or minor versions (MbedTLS anyone?), or bump the major version just for the sake of it even though there is no ABI/API breakage. This means that using compat bounds can both be too restrictive or not be helpful in preventing breakage. Basically, it's a mess. In #3024 I'm keeping a list of packages for which we know we really should specify compat bounds