Re: [Question] local paths when USE_THE_REPOSITORY_VARIABLE is not defined

[Patrick Steinhardt <ps@pks.im>]

On Mon, Oct 07, 2024 at 10:24:49PM +0530, Kousik Sanagavarapu wrote:
> Hi,
> 
> I have two questions but a bit of a background first -
> 
> 102de880d2 (path.c: migrate global git_path_* to take a repository
> argument, 2018-05-17) made global git_path_* functions take a repo
> argument.  The commit msg mentions that this migration doesn't change
> the local path functions in various builtins - which were defined using
> GIT_PATH_FUNC.  This was also the commit which introduced the macro
> REPO_GIT_PATH_FUNC.
> 
> Skip to 7ac16649ec (path: hide functions using `the_repository` by
> default, 2024-08-13), GIT_PATH_FUNC is hidden under
> USE_THE_REPOSITORY_VARIABLE and the REPO_GIT_PATH_FUNC is made its
> arbitrary repo equivalent - which can be inferred from the following
> portion of the diff
> 
> @@ -165,19 +130,10 @@ void report_linked_checkout_garbage(struct repository *r);
>  /*
>   * You can define a static memoized git path like:
>   *
> - *    static GIT_PATH_FUNC(git_path_foo, "FOO")
> + *    static REPO_GIT_PATH_FUNC(git_path_foo, "FOO")
>   *
>   * or use one of the global ones below.
>   */
> -#define GIT_PATH_FUNC(func, filename) \
> -       const char *func(void) \
> -       { \
> -               static char *ret; \
> -               if (!ret) \
> -                       ret = git_pathdup(filename); \
> -               return ret; \
> -       }
> -
>  #define REPO_GIT_PATH_FUNC(var, filename) \
>         const char *git_path_##var(struct repository *r) \
>         { \
> 
> (the GIT_PATH_FUNC macro is moved to be under USE_THE_REPOSITORY_VARIABLE)
> 
> Looking at the expansion of REPO_GIT_PATH_FUNC ...
> 
> #define REPO_GIT_PATH_FUNC(var, filename) \
> 	const char *git_path_##var(struct repository *r) \
> 	{ \
> 		if (!r->cached_paths.var) \
> 			r->cached_paths.var = repo_git_path(r, filename); \
> 		return r->cached_paths.var; \
> 	}
> 
> It seems that REPO_GIT_PATH_FUNC isn't an exact equivalent of
> GIT_PATH_FUNC.  That is, REPO_GIT_PATH_FUNC expects even a local path to be
> a field of the "struct repo_path_cache".  An example of a local path is
> EDIT_DESCRIPTION from "git branch --edit-description" (which inturn gets
> used by "git format-patch").
> 
> So my question is - do we want, in the future in which we are free from
> the dependency on "the_repository", for all the local paths to be a part
> of "struct repo_path_cache"?  Which in my gut feels wrong - one alternative
> then is that  we will have to refactor REPO_GIT_PATH_FUNC - or am I missing
> something here?

What I don't quite understand: what is the problem with making it part
of the `struct repo_path_cache`? Does this cause an actual issue, or is
it merely that you feel it is unnecessary complexity?

> I got into this when I was trying to refactor builtin/branch.c to be
> independent of "the_repository".  It was a very naive approach of just
> manual conversion of all the git_* calls to repo_* calls and similar
> changes but the compiler started to complain since I overlooked
> GIT_PATH_FUNC and some variables in environment.h which are also hidden
> under USE_THE_REPOSITORY_VARIABLE.
> 
> Which raises another question - why are variables such as
> "comment_line_str" and "default_abbrev" hidden under
> USE_THE_REPOSITORY_VARIABLE?[1]  They don't seem to be dependent on
> "the_repository"?  Again, I might be missing something here but am not
> sure what.

They do depend on `the_repository`, but implicitly only. The problem is
that those variables are populated via the config, and that may include
repository-local configuration. As such they contain values that have
been derived via `the_repository`, and those values may not be the
correct value when you handle multiple repositories in a single process,
because those may have a different value for e.g. "core.commentChar".

> By the way I don't expect this "naive approach" to be the right method
> of doing this - I was just tinkering to get to know
> USE_THE_REPOSITORY_VARIABLE better - since builtin/branch.c also calls
> into ref-filter which heavily relies on "the_repository" so changes
> there also would be appropriate for the complete picture.

Yeah, in the ideal case you'd first adapt any underlying code that you
happen to spot that relies on `the_repository`. That doesn't always
work as it is easy to miss that something implicitly depends on the
variable. But in case such a dependency is missed it will get to light
eventually as we continue with our quest to remove `the_repository`.

Patrick