Re: [PATCH] check_repository_format_version(): run git_default_config() again

From: Junio C Hamano <gitster@pobox.com>
Date: Aug 6, 2007, 20:26 UTC
Message-ID: <7vsl6wbe9e.fsf@assigned-by-dhcp.cox.net>
In-Reply-To: <7vbqdkcxy3.fsf@assigned-by-dhcp.cox.net>

Junio C Hamano <gitster@pobox.com> writes:

Show 7 quoted lines

> Sorry, I do not understand this patch from the commit log
> description.
>
> In the old days in 1.5.2 before either implementation of
> work-tree stuff, this function never called git_default_config()
> and presumably "git log" worked as expected for Brian.  Why is
> this call needed _here_, not in some of the callers?

Sorry scratch that.  v1.5.2 did not even have core.pager.

I see git_config(git_default_config) is called from setup_git_directory_gently() in the version that introduced core.pager. Curiously enough it does the config twice, once where no GIT_DIR is set and it sets inside_work_tree, and another by calling git_config(git_setup_config) -- big thanks for sorting this mess out.

I however have a mild suspicion that this has to be done much earlier.

For example, "git -p log" would first call handle_options(), which calls the setup_pager() logic, and then we call handle_internal_command() which calls setup_git_directory() that eventually leads to the callchain of finding the .git directory and .git/config file to be used. In that case, when we make the call to setup_pager() from handle_options(), we haven't even figured out if there is a configuration file, let alone reading from it.

We could work this around by having the "we need config -- where is it" logic in setup_pager(), but if we later have new options like -p that affects the way how git wrapper itself behaves based on the configuration, we would need the same "early config read" logic to support it.