[RFC] Batched low level index operations

首先阅读 PR #907 以及现有的 Index.Add() 和 Index.Remove() 入口点。将提议的 Index.Update/IndexUpdater API 与“repository 在调用之间保持一致”的要求进行比较；要视为完成，需要一个经 maintainer 批准的设计和明确的实现范围，而本 RFC 尚未提供这些内容。

In the light of https://github.com/libgit2/libgit2sharp/pull/907, I was wondering if we should take this approach one step further.

Each call to repo.Stage() will

• Perform a diff between the working directory and the Index in order to determine what additions/removals should be promoted to the staging area
• Perform additions/deletions in the in memory Index
• Eventually persist the in memory Index to the disk

Each call to repo.Index.Add() (or Remove()) will

• Perform an addition (or a deletion) in the in memory Index
• Persist the in memory Index to the disk

As such, invoking repo.Index.Add() or repo.Index.Remove() will be more efficient than a call to repo.Stage().

We want to allow the user to switch to the command line git.git, between each call to LibGit2Sharp, and find the repository in a coherent state. This is the requirement that compels us to persist the in-memory Index at the very end of each Index related method.

However, would clients of the library require to perform many sequential calls to the lower level methods repo.Index.Add()|Remove(), the Index would be persisted as part of each call. Which may not be very efficient.

Considering this above, would a need for an IndexUpdater exist?

We may expose something like in the Index type:

public virtual void Update(params Action<IndexUpdater>[] actions)

which would allow the client to write such code:

repo.Index.Update(i => i.Add(path), i => i.Remove(otherPath))

Or even some more elaborate custom thingies such as

private void StageAllAdditionsRemovalsAndModifications(IRepository repo)
{
    // Compare the Workdir against the Index without attempting at detecting Copies/Renames
    var changes = repo.Diff.Compare<TreeChanges>(null, true,
        compareOptions: new CompareOptions { Similarity = SimilarityOptions.None });

    repo.Index.Update(i =>
    {
        // We first clean the Index from unwanted entries...
        foreach (TreeEntryChanges treeEntryChanges in changes
            .Where(tec => tec.Status == ChangeKind.Deleted))
        {
            i.Remove(treeEntryChanges.Path);
        }

        // ...then insert in the Index the additions/modifications.
        foreach (TreeEntryChanges treeEntryChanges in changes
            .Where(tec => tec.Status == ChangeKind.Added || tec.Status == ChangeKind.Modified))
        {
            i.Add(treeEntryChanges.Path);
        }
    });
}

And allow us to only persist the Index only once, when all the actions have been performed by the IndexUpdater.

Thoughts?