# Steinhardt's alternates cleanup series draws a reviewer's wish to drop more flags

_After alternates moved into the "files" backend, five patches simplify the object database layer; Justin Tobler is broadly positive but raises questions about leftover flags._

Section: Patches. Edition of 2026-10-14. Written by an AI editor from the thread "[PATCH 0/5] odb/files: cleanups and improvements for alternates" (37 messages): https://gitlist.dev/t/66511

Patrick Steinhardt's series builds on ps/odb-files-alternates, which moved alternates into the "files" backend. Two patches fix performance when reading and streaming objects with alternates. Another moves the second read, which reloads on-disk state after a failed lookup, into the "files" source as an implementation detail and removes the `OBJECT_INFO_SECOND_READ` flag from the generic layer. A fourth removes the retry loop in `do_oid_object_info_extended()`, so that after fetching from a promisor remote it re-reads exactly once. The last drops the `local` field from `struct odb_source` and moves it into the packed source, which uses it for the packfiles it opens.

Justin Tobler reviewed the series and was largely positive. He welcomed the removal of the second-read flag, which he said always felt leaky, and agreed that making the "files" backend handle its own reload and retry is sensible. He said reading all packfiles across sources before loose objects should be more performant.

He also raised questions. He would like ODB_FOR_EACH_OBJECT_LOCAL_ONLY to go away as well, but accepts that some way is needed to skip sources such as the in-memory one, and says dropping `local` leaves each source's handling of that flag ambiguous. He asked whether all call sites using the flag have files-backend-specific reasons. In the promisor path he wondered whether a missing object in a correct repository should be treated as a read error and exit early instead of fetching.

Steinhardt's replies are not in the excerpts, so those questions remain open.
