Re: [PATCH 1/3] repository: move git_*_encoding configs to repo scope

From: Patrick Steinhardt <ps@pks.im>
Date: Oct 16, 2024, 06:15 UTC
Message-ID: <Zw9Z_xcHKcjKVUB4@pks.im>
In-Reply-To: <Zw6SsUyZ0oA0XqMK@ArchLinux>

On Wed, Oct 16, 2024 at 12:05:05AM +0800, shejialuo wrote:

Show 27 quoted lines

> There are many builtins will execute this config setups by calling
> "config.c::git_default_config" and then "git_default_i18n_config". If we
> need to use "repo" pointer, we may need to wrap this pointer. (This is
> not the problem and it is not hard).
> 
> But what if the "repo" pointer is NULL? We still need to set the value
> of these environment variables. Because when using "git-mailinfo(1)"
> outside of the repo, we still need to set "git_commit_encoding"
> according to the user's config.
> 
> So, from this perspective, I don't think it's a good idea to put these
> two configs into "struct repository". Because we can use these two
> configs outside of the repo, if we put them into "struct repository", it
> is strange.
> 
> However, I either don't know which way we would apply. So, I cannot give
> accurate answer here.
> 
> ---
> 
> Patrick, I wanna ask you a question here. What's your envision here in
> above situation. As you can see, if we put some configs into "struct
> repository" and we run the builtins outside of the repo where we need to
> set up configs, the "repo" is NULL. And we will get into trouble.
> 
> My idea is that if a config could be used outside of the repo, then we
> should not put it into "struct repository".

I guess it would be nice to have a set of convenice functions in our config code that know to handle the case where the passed-in repository is `NULL`. If so, they'd only parse the global- and system-level config. If it is set, it would parse all three of them.

I also kind of agree that it should likely not be put into the `struct repository` in that case. I think where exactly to put things will always depend on the current usecase. I bet that in most cases, we should be able to get away with not storing the value anywhere global at all, which would be the best solution in my opinion:

  - It can either be stored on the stack if we don't have to pass it
    around everywhere.

  - It can be passed around in a specific structure if we pass the value
    within in a certain subsystem, only.

  - Or we can parse it on an as-needed basis if it happens deep down in
    the calling stack when it's used essentially everwhere.

Now there will be situations where we used to store things globally as a caching mechanism, and not caching it may have performance impacts. But I guess that most cases do not fall into this category.

Patrick