Colin Hinton's patch changes fetch.followRemoteHEAD handling. The value was validated while config was parsed, producing a warning even when the fetch would never use it. The patch stores the raw string and resolves it at the one place in do_fetch() that needs it, so a mistyped value only warns, and a missing value only dies, when the fetch would have consulted it.

The per-remote remote.<name>.followRemoteHEAD setting in remote.c's handle_config() has the same problem. Hinton left it alone because it touches shared remote-parsing code, and added a NEEDSWORK comment there.

Junio Hamano said only two minor things remain in this iteration. One is a commit message wording fix: there is a single NEEDSWORK comment, not one "at both" places. The other concerned the initialization of the config struct in cmd_fetch().

Matt Hunter, who had no time to test the new revision, noted that the new comment in fetch.c does not carry the NEEDSWORK label and suggested adding it rather than rewording the sentence. He also reasoned about the new follow_remote_head_seen flag: when it is true, the raw string is NULL for a valueless true, empty for an explicitly empty value, and otherwise the normal value, which affects how one case is handled.

Separately, Hamano mentioned in the thread on Harald Nordgren's fetch.shallow patch that it textually conflicts with this work and that he resolved the conflicts when merging both to 'seen'.