The original patch added fetch.shallow, off by default. With it, a fetch or pull in a shallow repository not already scoped to specific refs fetches only the current branch's tracked upstream instead of every remote branch. The motivation was that plain git pull in a shallow, sparsely checked-out clone of a repository with many branches can take minutes or hang.
The thread now carries a larger series. One patch makes remote.<name>.fetchRefmap work without remote.<name>.fetch, inferring what to fetch from local branches whose upstream is on that remote. The refmap only controls where fetched refs land. Another patch adds a --limited-fetch option to git remote add that sets up fetchRefmap instead of a wildcard fetch refspec. A third turns it on by default when the repository is already shallow and neither -t/--track nor --mirror is given. A refactor extracts the check for whether a branch builds on a remote.
Junio C Hamano said remote.cabal.fetch should behave as if its refnames were written on the command line when nothing is given there. Refspecs without a destination side store results only in FETCH_HEAD, so a git pull can still create an octopus merge. He hoped refmap would let users say where to store them. Phillip Wood said it is unfortunate this needs separate refspecs to reject non-fast-forward updates of some branches while mapping refs the same way.
Wood also asked whether setting a refmap alongside the fetch refspec would make older Git versions try to update local branches when fetching. He suggested a separate config setting for "what to fetch", so users could unset remote.<name>.fetch and older versions would fail rather than overwrite local branches.
D. Ben Knoble said he was unsure the proposed wording captured a "narrow refspec" remote. After reading the new version he said he sees what Nordgren is after, and asked for an example of adding a remote, since he wants big fetches from his primary remotes. The design question of how the new settings interact with fetch refspecs is unresolved.