Adam Majer has been experimenting with keeping both OID types in a single repository. A first attempt with nodejs failed because of missing submodules. He then tried the Linux tree, which proved more problematic.

He uses convert_commit_object to convert a commit from one OID type to the other. The current implementation has comments about needing to convert every kind of OID reference, including mergetag, and it bails out on other, unknown headers. In the Linux tree, about a dozen mergetags contain references that do not appear to be fetchable from anywhere. He also says a number of commits in Linux history carry headers the function does not recognize.

In a follow-up, Majer said he tried the other repositories the mergetags refer to, but the objects stayed unresolved. One mergetag pointed to a third-party URL that did not work over https, but http:// access resolved it, so one of the dozen was found.

Majer argued that conversion need not resolve those objects. Any supported hash function should be unique enough that all of them can be tried, so mergetags and proprietary headers could be treated as text in the conversion function and retrieved later using the original hash function. In his view the hash function does not affect their existence or validity.

brian m. carlson also took part in the thread, but his remarks are not in the excerpts, so it is unresolved whether the project would accept treating such headers as opaque text.